Fyxer uses OpenAI models, fine-tuning, memory, and real user feedback to organize inboxes and draft emails in each user's own voice.
Fyxer uses OpenAI models, fine-tuning, memory, and real user feedback to organize inboxes and draft emails in each user's own voice.
RSC Advances Using Capsicum annuum L. (red pepper) seed extract as a natural reducing and capping agent, this study green-synthesized Ag/MgO nanocomposites, systematically evaluated sunlight-driven diclofenac degradation across irradiation time, catalyst dosage, Ag loading, pH, initial DCF concentration, electrolyte type and concentration, and irradiation source, found that 10% Ag/MgO was most active, achieving 80% degradation of 10 mg/L diclofenac within 180 min with hydroxyl radicals identified as the dominant reactive species, and developed an EEFO-optimized DT_LSBoost model (R = 0.9999; RMSE = 1.2 × 10⁻³) with a MATLAB graphical interface for rapid prediction.
Using Capsicum annuum L. (red pepper) seed extract as a natural reducing and capping agent, this study green-synthesized Ag/MgO nanocomposites, systematically evaluated sunlight-driven diclofenac degradation across irradiation time, catalyst dosage, Ag loading, pH, initial DCF concentration, electrolyte type and concentration, and irradiation source, found that 10% Ag/MgO was most active, achieving 80% degradation of 10 mg/L diclofenac within 180 min with hydroxyl radicals identified as the dominant reactive species, and developed an EEFO-optimized DT_LSBoost model (R = 0.9999; RMSE = 1.2 × 10⁻³) with a MATLAB graphical interface for rapid prediction.
Using Capsicum annuum L. (red pepper) seed extract as a natural reducing and capping agent, this study green-synthesized Ag/MgO nanocomposites, systematically evaluated sunlight-driven diclofenac degradation across irradiation time, catalyst dosage, Ag loading, pH, initial DCF concentration, electrolyte type and concentration, and irradiation source, found that 10% Ag/MgO was most active, achieving 80% degradation of 10 mg/L diclofenac within 180 min with hydroxyl radicals identified as the dominant reactive species, and developed an EEFO-optimized DT_LSBoost model (R = 0.9999; RMSE = 1.2 × 10⁻³) with a MATLAB graphical interface for rapid prediction.
Using Capsicum annuum L. (red pepper) seed extract as a natural reducing and capping agent, this study green-synthesized Ag/MgO nanocomposites, systematically evaluated sunlight-driven diclofenac degradation across irradiation time, catalyst dosage, Ag loading, pH, initial DCF concentration, electrolyte type and concentration, and irradiation source, found that 10% Ag/MgO was most active, achieving 80% degradation of 10 mg/L diclofenac within 180 min with hydroxyl radicals identified as the dominant reactive species, and developed an EEFO-optimized DT_LSBoost model (R = 0.9999; RMSE = 1.2 × 10⁻³) with a MATLAB graphical interface for rapid prediction.
Surgical infections This article reviews evidence and implementation experience on surgical site infection (SSI) prevention across infection prevention and control (IPC), antimicrobial stewardship (AMS), antimicrobial resistance (AMR), quality improvement, surveillance, safety culture, human factors, and emerging technologies, concluding that multimodal, context-adapted IPC interventions can substantially reduce SSI incidence without major capital investment, that persistent gaps remain in infrastructure, guideline implementation, surveillance, and environmental controls, that AMR further emphasizes the need to integrate AMS into surgical safety frameworks, and that artificial intelligence and objective diagnostic technologies may improve surveillance and early SSI detection, so that SSI prevention is both a
This article reviews evidence and implementation experience on surgical site infection (SSI) prevention across infection prevention and control (IPC), antimicrobial stewardship (AMS), antimicrobial resistance (AMR), quality improvement, surveillance, safety culture, human factors, and emerging technologies, concluding that multimodal, context-adapted IPC interventions can substantially reduce SSI incidence without major capital investment, that persistent gaps remain in infrastructure, guideline implementation, surveillance, and environmental controls, that AMR further emphasizes the need to integrate AMS into surgical safety frameworks, and that artificial intelligence and objective diagnostic technologies may improve surveillance and early SSI detection, so that SSI prevention is both a
This article reviews evidence and implementation experience on surgical site infection (SSI) prevention across infection prevention and control (IPC), antimicrobial stewardship (AMS), antimicrobial resistance (AMR), quality improvement, surveillance, safety culture, human factors, and emerging technologies, concluding that multimodal, context-adapted IPC interventions can substantially reduce SSI incidence without major capital investment, that persistent gaps remain in infrastructure, guideline implementation, surveillance, and environmental controls, that AMR further emphasizes the need to integrate AMS into surgical safety frameworks, and that artificial intelligence and objective diagnostic technologies may improve surveillance and early SSI detection, so that SSI prevention is both a
This article reviews evidence and implementation experience on surgical site infection (SSI) prevention across infection prevention and control (IPC), antimicrobial stewardship (AMS), antimicrobial resistance (AMR), quality improvement, surveillance, safety culture, human factors, and emerging technologies, concluding that multimodal, context-adapted IPC interventions can substantially reduce SSI incidence without major capital investment, that persistent gaps remain in infrastructure, guideline implementation, surveillance, and environmental controls, that AMR further emphasizes the need to integrate AMS into surgical safety frameworks, and that artificial intelligence and objective diagnostic technologies may improve surveillance and early SSI detection, so that SSI prevention is both a
Journal of Chemical Information and Modeling Using UV absorption wavelength (λmax) prediction as a testbed across 18,415 solute-solvent pairs (after the Greenman/Song duplicate-handling protocol) with stratified 5-fold cross-validation and two external data sets totaling over 40,000 molecules, this work compares five models spanning four architecture families (Random Forest, XGBoost, a directed message-passing GNN, a BiGRU, and a pretrained Transformer) and finds that Random Forest is statistically indistinguishable from the best deep learning model for screening within known chemical space (RMSE 31.50 ± 1.47 vs 33.15 ± 3.27 nm, p = 0.16) while training in 15 min on a CPU, whereas on 16 novel UV-absorbing compounds the GNN (MAE 26.2 nm), Transformer (26.3 nm), and BiGRU (28.6 nm) all outperform RF (38.
Using UV absorption wavelength (λmax) prediction as a testbed across 18,415 solute-solvent pairs (after the Greenman/Song duplicate-handling protocol) with stratified 5-fold cross-validation and two external data sets totaling over 40,000 molecules, this work compares five models spanning four architecture families (Random Forest, XGBoost, a directed message-passing GNN, a BiGRU, and a pretrained Transformer) and finds that Random Forest is statistically indistinguishable from the best deep learning model for screening within known chemical space (RMSE 31.50 ± 1.47 vs 33.15 ± 3.27 nm, p = 0.16) while training in 15 min on a CPU, whereas on 16 novel UV-absorbing compounds the GNN (MAE 26.2 nm), Transformer (26.3 nm), and BiGRU (28.6 nm) all outperform RF (38.
Using UV absorption wavelength (λmax) prediction as a testbed across 18,415 solute-solvent pairs (after the Greenman/Song duplicate-handling protocol) with stratified 5-fold cross-validation and two external data sets totaling over 40,000 molecules, this work compares five models spanning four architecture families (Random Forest, XGBoost, a directed message-passing GNN, a BiGRU, and a pretrained Transformer) and finds that Random Forest is statistically indistinguishable from the best deep learning model for screening within known chemical space (RMSE 31.50 ± 1.47 vs 33.15 ± 3.27 nm, p = 0.16) while training in 15 min on a CPU, whereas on 16 novel UV-absorbing compounds the GNN (MAE 26.2 nm), Transformer (26.3 nm), and BiGRU (28.6 nm) all outperform RF (38.
Using UV absorption wavelength (λmax) prediction as a testbed across 18,415 solute-solvent pairs (after the Greenman/Song duplicate-handling protocol) with stratified 5-fold cross-validation and two external data sets totaling over 40,000 molecules, this work compares five models spanning four architecture families (Random Forest, XGBoost, a directed message-passing GNN, a BiGRU, and a pretrained Transformer) and finds that Random Forest is statistically indistinguishable from the best deep learning model for screening within known chemical space (RMSE 31.50 ± 1.47 vs 33.15 ± 3.27 nm, p = 0.16) while training in 15 min on a CPU, whereas on 16 novel UV-absorbing compounds the GNN (MAE 26.2 nm), Transformer (26.3 nm), and BiGRU (28.6 nm) all outperform RF (38.
Journal of Chemical Information and Modeling This study evaluates state-of-the-art machine-learning-based methods for nanobody structure prediction and benchmarks various HADDOCK workflows for modeling nanobody-antigen interactions across different input nanobody ensembles and information scenarios, proposing an ensemble docking pipeline that starts from nanobody structural models predicted by AlphaFold2 and ImmuneBuilder and, provided some epitope information is available, achieves higher success rates than the AlphaFold baseline on all generated models.
This study evaluates state-of-the-art machine-learning-based methods for nanobody structure prediction and benchmarks various HADDOCK workflows for modeling nanobody-antigen interactions across different input nanobody ensembles and information scenarios, proposing an ensemble docking pipeline that starts from nanobody structural models predicted by AlphaFold2 and ImmuneBuilder and, provided some epitope information is available, achieves higher success rates than the AlphaFold baseline on all generated models.
This study evaluates state-of-the-art machine-learning-based methods for nanobody structure prediction and benchmarks various HADDOCK workflows for modeling nanobody-antigen interactions across different input nanobody ensembles and information scenarios, proposing an ensemble docking pipeline that starts from nanobody structural models predicted by AlphaFold2 and ImmuneBuilder and, provided some epitope information is available, achieves higher success rates than the AlphaFold baseline on all generated models.
This study evaluates state-of-the-art machine-learning-based methods for nanobody structure prediction and benchmarks various HADDOCK workflows for modeling nanobody-antigen interactions across different input nanobody ensembles and information scenarios, proposing an ensemble docking pipeline that starts from nanobody structural models predicted by AlphaFold2 and ImmuneBuilder and, provided some epitope information is available, achieves higher success rates than the AlphaFold baseline on all generated models.
Communications Psychology This Perspective argues that generative AI should not be understood as just another teaching tool for efficiency and personalization, but as an active, persistent, generalist, and increasingly autonomous cognitive artefact and 'epistemic infrastructure' that restructures schools' cognitive ecologies by redistributing epistemic labour and consolidating feedback, assessment, explanation, content generation, and tutoring within a single interface; on this basis the authors identify two enduring misalignments, a pedagogical gap between learning sciences and AI design and a goal gap between measurable performance and developmental aims, both reflecting logics of efficiency, standardization, and control already embedded in existing educational systems, which GenAI does not introduce but risks en
This Perspective argues that generative AI should not be understood as just another teaching tool for efficiency and personalization, but as an active, persistent, generalist, and increasingly autonomous cognitive artefact and 'epistemic infrastructure' that restructures schools' cognitive ecologies by redistributing epistemic labour and consolidating feedback, assessment, explanation, content generation, and tutoring within a single interface; on this basis the authors identify two enduring misalignments, a pedagogical gap between learning sciences and AI design and a goal gap between measurable performance and developmental aims, both reflecting logics of efficiency, standardization, and control already embedded in existing educational systems, which GenAI does not introduce but risks en
This Perspective argues that generative AI should not be understood as just another teaching tool for efficiency and personalization, but as an active, persistent, generalist, and increasingly autonomous cognitive artefact and 'epistemic infrastructure' that restructures schools' cognitive ecologies by redistributing epistemic labour and consolidating feedback, assessment, explanation, content generation, and tutoring within a single interface; on this basis the authors identify two enduring misalignments, a pedagogical gap between learning sciences and AI design and a goal gap between measurable performance and developmental aims, both reflecting logics of efficiency, standardization, and control already embedded in existing educational systems, which GenAI does not introduce but risks en
This Perspective argues that generative AI should not be understood as just another teaching tool for efficiency and personalization, but as an active, persistent, generalist, and increasingly autonomous cognitive artefact and 'epistemic infrastructure' that restructures schools' cognitive ecologies by redistributing epistemic labour and consolidating feedback, assessment, explanation, content generation, and tutoring within a single interface; on this basis the authors identify two enduring misalignments, a pedagogical gap between learning sciences and AI design and a goal gap between measurable performance and developmental aims, both reflecting logics of efficiency, standardization, and control already embedded in existing educational systems, which GenAI does not introduce but risks en
arXiv Through dual representation space analysis, this work identifies an "intermediate laundering region" mechanism, builds the PADBen benchmark with a five-type text taxonomy and five progressive detection tasks, and evaluates 11 detectors, revealing a critical asymmetry: paraphrase attacks do not universally defeat detection—plagiarism evasion (paraphrasing LLM-generated text) remains detectable (RADAR sentence-pair AUC 0.909), while authorship obfuscation (paraphrasing human-authored text) collapses detection to near-random performance (AUC 0.526–0.748).
Through dual representation space analysis, this work identifies an "intermediate laundering region" mechanism, builds the PADBen benchmark with a five-type text taxonomy and five progressive detection tasks, and evaluates 11 detectors, revealing a critical asymmetry: paraphrase attacks do not universally defeat detection—plagiarism evasion (paraphrasing LLM-generated text) remains detectable (RADAR sentence-pair AUC 0.909), while authorship obfuscation (paraphrasing human-authored text) collapses detection to near-random performance (AUC 0.526–0.748).
Through dual representation space analysis, this work identifies an "intermediate laundering region" mechanism, builds the PADBen benchmark with a five-type text taxonomy and five progressive detection tasks, and evaluates 11 detectors, revealing a critical asymmetry: paraphrase attacks do not universally defeat detection—plagiarism evasion (paraphrasing LLM-generated text) remains detectable (RADAR sentence-pair AUC 0.909), while authorship obfuscation (paraphrasing human-authored text) collapses detection to near-random performance (AUC 0.526–0.748).
Through dual representation space analysis, this work identifies an "intermediate laundering region" mechanism, builds the PADBen benchmark with a five-type text taxonomy and five progressive detection tasks, and evaluates 11 detectors, revealing a critical asymmetry: paraphrase attacks do not universally defeat detection—plagiarism evasion (paraphrasing LLM-generated text) remains detectable (RADAR sentence-pair AUC 0.909), while authorship obfuscation (paraphrasing human-authored text) collapses detection to near-random performance (AUC 0.526–0.748).
发表出处待核验 This work builds a frozen, content-hashed data asset and a uniform scoring harness that maps seven MaveDB saturation genome editing (SGE) score sets to GRCh38, harmonises orientation and freezes them into immutable matrices, then evaluates nineteen variant effect predictors across sixteen strata using per-gene Spearman rho pooled by DerSimonian-Laird random-effects meta-analysis with per-stratum measurement-reliability estimates, attenuation correction, paired dependent-correlation tests and leave-one-gene-out validation, covering 64,178 variants and seven cancer susceptibility genes.
This work builds a frozen, content-hashed data asset and a uniform scoring harness that maps seven MaveDB saturation genome editing (SGE) score sets to GRCh38, harmonises orientation and freezes them into immutable matrices, then evaluates nineteen variant effect predictors across sixteen strata using per-gene Spearman rho pooled by DerSimonian-Laird random-effects meta-analysis with per-stratum measurement-reliability estimates, attenuation correction, paired dependent-correlation tests and leave-one-gene-out validation, covering 64,178 variants and seven cancer susceptibility genes.
This work builds a frozen, content-hashed data asset and a uniform scoring harness that maps seven MaveDB saturation genome editing (SGE) score sets to GRCh38, harmonises orientation and freezes them into immutable matrices, then evaluates nineteen variant effect predictors across sixteen strata using per-gene Spearman rho pooled by DerSimonian-Laird random-effects meta-analysis with per-stratum measurement-reliability estimates, attenuation correction, paired dependent-correlation tests and leave-one-gene-out validation, covering 64,178 variants and seven cancer susceptibility genes.
This work builds a frozen, content-hashed data asset and a uniform scoring harness that maps seven MaveDB saturation genome editing (SGE) score sets to GRCh38, harmonises orientation and freezes them into immutable matrices, then evaluates nineteen variant effect predictors across sixteen strata using per-gene Spearman rho pooled by DerSimonian-Laird random-effects meta-analysis with per-stratum measurement-reliability estimates, attenuation correction, paired dependent-correlation tests and leave-one-gene-out validation, covering 64,178 variants and seven cancer susceptibility genes.
Advances in wound care In a single-center retrospective cohort of 584 adults first admitted for a chronic wound between 2021 and 2024, a two-variable model combining the admission-day red cell distribution width-albumin ratio (RAR) with the age-adjusted Charlson Comorbidity Index (ACCI) predicted 1-year all-cause mortality with an optimism-corrected C-statistic of 0.855 and good calibration, stratifying patients into low-, intermediate-, and high-risk tiers (observed mortality 1.3%, 9.0%, and 36.8%), while discrimination was similar but calibration imprecise in an exploratory temporal evaluation of a later same-center cohort of 124 patients.
In a single-center retrospective cohort of 584 adults first admitted for a chronic wound between 2021 and 2024, a two-variable model combining the admission-day red cell distribution width-albumin ratio (RAR) with the age-adjusted Charlson Comorbidity Index (ACCI) predicted 1-year all-cause mortality with an optimism-corrected C-statistic of 0.855 and good calibration, stratifying patients into low-, intermediate-, and high-risk tiers (observed mortality 1.3%, 9.0%, and 36.8%), while discrimination was similar but calibration imprecise in an exploratory temporal evaluation of a later same-center cohort of 124 patients.
In a single-center retrospective cohort of 584 adults first admitted for a chronic wound between 2021 and 2024, a two-variable model combining the admission-day red cell distribution width-albumin ratio (RAR) with the age-adjusted Charlson Comorbidity Index (ACCI) predicted 1-year all-cause mortality with an optimism-corrected C-statistic of 0.855 and good calibration, stratifying patients into low-, intermediate-, and high-risk tiers (observed mortality 1.3%, 9.0%, and 36.8%), while discrimination was similar but calibration imprecise in an exploratory temporal evaluation of a later same-center cohort of 124 patients.
In a single-center retrospective cohort of 584 adults first admitted for a chronic wound between 2021 and 2024, a two-variable model combining the admission-day red cell distribution width-albumin ratio (RAR) with the age-adjusted Charlson Comorbidity Index (ACCI) predicted 1-year all-cause mortality with an optimism-corrected C-statistic of 0.855 and good calibration, stratifying patients into low-, intermediate-, and high-risk tiers (observed mortality 1.3%, 9.0%, and 36.8%), while discrimination was similar but calibration imprecise in an exploratory temporal evaluation of a later same-center cohort of 124 patients.
Nature Communications The work introduces exponentially tilted thermodynamic maps (expTM), which add an exponential tilting factor to the Gaussian prior of thermodynamic maps so that the prior mean and variance correspond to pressure or chemical potential and to temperature, enabling thermodynamically correct sampling at arbitrary temperature, pressure, or chemical potential from only a few observations far from the phase boundary; the authors reproduce the lattice-gas phase transition in the grand canonical ensemble using two data points (density difference within 0.05 except near the critical chemical potential) and predict CO2 phase transitions under varying pressure in the isothermal-isobaric ensemble, identifying an intermediate state between Phase I and Phase III at roughly 4.5–5.4 GPa.
The work introduces exponentially tilted thermodynamic maps (expTM), which add an exponential tilting factor to the Gaussian prior of thermodynamic maps so that the prior mean and variance correspond to pressure or chemical potential and to temperature, enabling thermodynamically correct sampling at arbitrary temperature, pressure, or chemical potential from only a few observations far from the phase boundary; the authors reproduce the lattice-gas phase transition in the grand canonical ensemble using two data points (density difference within 0.05 except near the critical chemical potential) and predict CO2 phase transitions under varying pressure in the isothermal-isobaric ensemble, identifying an intermediate state between Phase I and Phase III at roughly 4.5–5.4 GPa.
The work introduces exponentially tilted thermodynamic maps (expTM), which add an exponential tilting factor to the Gaussian prior of thermodynamic maps so that the prior mean and variance correspond to pressure or chemical potential and to temperature, enabling thermodynamically correct sampling at arbitrary temperature, pressure, or chemical potential from only a few observations far from the phase boundary; the authors reproduce the lattice-gas phase transition in the grand canonical ensemble using two data points (density difference within 0.05 except near the critical chemical potential) and predict CO2 phase transitions under varying pressure in the isothermal-isobaric ensemble, identifying an intermediate state between Phase I and Phase III at roughly 4.5–5.4 GPa.
The work introduces exponentially tilted thermodynamic maps (expTM), which add an exponential tilting factor to the Gaussian prior of thermodynamic maps so that the prior mean and variance correspond to pressure or chemical potential and to temperature, enabling thermodynamically correct sampling at arbitrary temperature, pressure, or chemical potential from only a few observations far from the phase boundary; the authors reproduce the lattice-gas phase transition in the grand canonical ensemble using two data points (density difference within 0.05 except near the critical chemical potential) and predict CO2 phase transitions under varying pressure in the isothermal-isobaric ensemble, identifying an intermediate state between Phase I and Phase III at roughly 4.5–5.4 GPa.
Journal of Chemical Information and Modeling Pro-GAT is a graph attention-based framework for geometry-preserving molecular graph repair that operates on chemically disconnected diffusion-generated PROTAC candidates by predicting bounded coordinate corrections and constrained atom-type modifications through geometry-aware graph attention layers; it recovers 31.58% of disconnected structures in the DiffPROTAC test set, and when combined with DiffPROTAC and DiffLinker fine-tuned on PROTAC-specific data, it improves the proportion of chemically valid candidates in the aggregated output from 76.70% to 83.92% and from 63.16% to 68.73%, while maintaining uniqueness levels of 80.18% and 63.
Pro-GAT is a graph attention-based framework for geometry-preserving molecular graph repair that operates on chemically disconnected diffusion-generated PROTAC candidates by predicting bounded coordinate corrections and constrained atom-type modifications through geometry-aware graph attention layers; it recovers 31.58% of disconnected structures in the DiffPROTAC test set, and when combined with DiffPROTAC and DiffLinker fine-tuned on PROTAC-specific data, it improves the proportion of chemically valid candidates in the aggregated output from 76.70% to 83.92% and from 63.16% to 68.73%, while maintaining uniqueness levels of 80.18% and 63.
Pro-GAT is a graph attention-based framework for geometry-preserving molecular graph repair that operates on chemically disconnected diffusion-generated PROTAC candidates by predicting bounded coordinate corrections and constrained atom-type modifications through geometry-aware graph attention layers; it recovers 31.58% of disconnected structures in the DiffPROTAC test set, and when combined with DiffPROTAC and DiffLinker fine-tuned on PROTAC-specific data, it improves the proportion of chemically valid candidates in the aggregated output from 76.70% to 83.92% and from 63.16% to 68.73%, while maintaining uniqueness levels of 80.18% and 63.
Pro-GAT is a graph attention-based framework for geometry-preserving molecular graph repair that operates on chemically disconnected diffusion-generated PROTAC candidates by predicting bounded coordinate corrections and constrained atom-type modifications through geometry-aware graph attention layers; it recovers 31.58% of disconnected structures in the DiffPROTAC test set, and when combined with DiffPROTAC and DiffLinker fine-tuned on PROTAC-specific data, it improves the proportion of chemically valid candidates in the aggregated output from 76.70% to 83.92% and from 63.16% to 68.73%, while maintaining uniqueness levels of 80.18% and 63.
Journal of Chemical Information and Modeling This Perspective argues that chemistry and language have followed a comparable representational trajectory—from tacit practitioner knowledge to systematic symbols and then to geometric, navigable spaces—and builds on this to propose a conceptual framework in which molecular properties are interpreted as forms of chemical meaning emerging from molecular structure, treating semantic interpolation in latent spaces and alchemical transformations in molecular simulation as related strategies for navigating continuous manifolds that connect discrete chemical states, while proposing a Chemical Imitation Game as a possible framework for evaluating molecular AI beyond syntactic validity toward chemically meaningful reasoning.
This Perspective argues that chemistry and language have followed a comparable representational trajectory—from tacit practitioner knowledge to systematic symbols and then to geometric, navigable spaces—and builds on this to propose a conceptual framework in which molecular properties are interpreted as forms of chemical meaning emerging from molecular structure, treating semantic interpolation in latent spaces and alchemical transformations in molecular simulation as related strategies for navigating continuous manifolds that connect discrete chemical states, while proposing a Chemical Imitation Game as a possible framework for evaluating molecular AI beyond syntactic validity toward chemically meaningful reasoning.
This Perspective argues that chemistry and language have followed a comparable representational trajectory—from tacit practitioner knowledge to systematic symbols and then to geometric, navigable spaces—and builds on this to propose a conceptual framework in which molecular properties are interpreted as forms of chemical meaning emerging from molecular structure, treating semantic interpolation in latent spaces and alchemical transformations in molecular simulation as related strategies for navigating continuous manifolds that connect discrete chemical states, while proposing a Chemical Imitation Game as a possible framework for evaluating molecular AI beyond syntactic validity toward chemically meaningful reasoning.
This Perspective argues that chemistry and language have followed a comparable representational trajectory—from tacit practitioner knowledge to systematic symbols and then to geometric, navigable spaces—and builds on this to propose a conceptual framework in which molecular properties are interpreted as forms of chemical meaning emerging from molecular structure, treating semantic interpolation in latent spaces and alchemical transformations in molecular simulation as related strategies for navigating continuous manifolds that connect discrete chemical states, while proposing a Chemical Imitation Game as a possible framework for evaluating molecular AI beyond syntactic validity toward chemically meaningful reasoning.
Journal of Chemical Information and Modeling This work introduces PockLigGPT, a GPT-based framework for ligand generation conditioned on the amino acid sequence of a protein pocket, trained in four stages (large-scale ZINC20 chemical pretraining, ChEMBL bioactivity-oriented adaptation, pocket-sequence-conditioned fine-tuning, and pocket-specific docking-guided reinforcement learning with AutoDock Vina-based rewards), achieving competitive docking-oriented performance under a standardized evaluation protocol while maintaining chemical plausibility and Lipinski-based drug-likeness, with docking studies on Alzheimer's disease-associated targets and token-level analyses supporting its utility for de novo drug design.
This work introduces PockLigGPT, a GPT-based framework for ligand generation conditioned on the amino acid sequence of a protein pocket, trained in four stages (large-scale ZINC20 chemical pretraining, ChEMBL bioactivity-oriented adaptation, pocket-sequence-conditioned fine-tuning, and pocket-specific docking-guided reinforcement learning with AutoDock Vina-based rewards), achieving competitive docking-oriented performance under a standardized evaluation protocol while maintaining chemical plausibility and Lipinski-based drug-likeness, with docking studies on Alzheimer's disease-associated targets and token-level analyses supporting its utility for de novo drug design.
This work introduces PockLigGPT, a GPT-based framework for ligand generation conditioned on the amino acid sequence of a protein pocket, trained in four stages (large-scale ZINC20 chemical pretraining, ChEMBL bioactivity-oriented adaptation, pocket-sequence-conditioned fine-tuning, and pocket-specific docking-guided reinforcement learning with AutoDock Vina-based rewards), achieving competitive docking-oriented performance under a standardized evaluation protocol while maintaining chemical plausibility and Lipinski-based drug-likeness, with docking studies on Alzheimer's disease-associated targets and token-level analyses supporting its utility for de novo drug design.
This work introduces PockLigGPT, a GPT-based framework for ligand generation conditioned on the amino acid sequence of a protein pocket, trained in four stages (large-scale ZINC20 chemical pretraining, ChEMBL bioactivity-oriented adaptation, pocket-sequence-conditioned fine-tuning, and pocket-specific docking-guided reinforcement learning with AutoDock Vina-based rewards), achieving competitive docking-oriented performance under a standardized evaluation protocol while maintaining chemical plausibility and Lipinski-based drug-likeness, with docking studies on Alzheimer's disease-associated targets and token-level analyses supporting its utility for de novo drug design.
RSC Advances This review surveys how machine learning and artificial intelligence, combined with high-throughput first-principles calculations and experimental databases, are used to predict key spintronic magnetic properties (such as Curie temperature, magnetocrystalline anisotropy energy, magnetic moment, spin Hall conductivity, Dzyaloshinskii–Moriya interaction, coercivity, and tunneling magnetoresistance), discusses descriptor engineering, graph neural networks, and physics-informed learning, reviews AI-assisted screening and inverse design across Heusler alloys, topological spin systems, and two-dimensional magnets, and outlines a future roadmap toward physics-guided, uncertainty-aware, and autonomous closed-loop optimization.
This review surveys how machine learning and artificial intelligence, combined with high-throughput first-principles calculations and experimental databases, are used to predict key spintronic magnetic properties (such as Curie temperature, magnetocrystalline anisotropy energy, magnetic moment, spin Hall conductivity, Dzyaloshinskii–Moriya interaction, coercivity, and tunneling magnetoresistance), discusses descriptor engineering, graph neural networks, and physics-informed learning, reviews AI-assisted screening and inverse design across Heusler alloys, topological spin systems, and two-dimensional magnets, and outlines a future roadmap toward physics-guided, uncertainty-aware, and autonomous closed-loop optimization.
This review surveys how machine learning and artificial intelligence, combined with high-throughput first-principles calculations and experimental databases, are used to predict key spintronic magnetic properties (such as Curie temperature, magnetocrystalline anisotropy energy, magnetic moment, spin Hall conductivity, Dzyaloshinskii–Moriya interaction, coercivity, and tunneling magnetoresistance), discusses descriptor engineering, graph neural networks, and physics-informed learning, reviews AI-assisted screening and inverse design across Heusler alloys, topological spin systems, and two-dimensional magnets, and outlines a future roadmap toward physics-guided, uncertainty-aware, and autonomous closed-loop optimization.
This review surveys how machine learning and artificial intelligence, combined with high-throughput first-principles calculations and experimental databases, are used to predict key spintronic magnetic properties (such as Curie temperature, magnetocrystalline anisotropy energy, magnetic moment, spin Hall conductivity, Dzyaloshinskii–Moriya interaction, coercivity, and tunneling magnetoresistance), discusses descriptor engineering, graph neural networks, and physics-informed learning, reviews AI-assisted screening and inverse design across Heusler alloys, topological spin systems, and two-dimensional magnets, and outlines a future roadmap toward physics-guided, uncertainty-aware, and autonomous closed-loop optimization.
International Journal of Behavioral Medicine This cross-sectional study surveyed 427 adults on antiretroviral treatment (ART) at a large tertiary referral center in Kano, Nigeria, finding that 75.2% were aware of AI chatbots, 72.6% had ever used one, and 66.5% had used chatbots for HIV-related queries (most commonly general HIV/ART information 36.3%, checking ART side effects 23.4%, and preparing questions for clinicians 15.0%), and identified factors independently associated with HIV-related chatbot use, including younger age, post-secondary education, being married, shorter ART duration, presence of comorbidities, smartphone ownership, internet access, and English proficiency.
This cross-sectional study surveyed 427 adults on antiretroviral treatment (ART) at a large tertiary referral center in Kano, Nigeria, finding that 75.2% were aware of AI chatbots, 72.6% had ever used one, and 66.5% had used chatbots for HIV-related queries (most commonly general HIV/ART information 36.3%, checking ART side effects 23.4%, and preparing questions for clinicians 15.0%), and identified factors independently associated with HIV-related chatbot use, including younger age, post-secondary education, being married, shorter ART duration, presence of comorbidities, smartphone ownership, internet access, and English proficiency.
This cross-sectional study surveyed 427 adults on antiretroviral treatment (ART) at a large tertiary referral center in Kano, Nigeria, finding that 75.2% were aware of AI chatbots, 72.6% had ever used one, and 66.5% had used chatbots for HIV-related queries (most commonly general HIV/ART information 36.3%, checking ART side effects 23.4%, and preparing questions for clinicians 15.0%), and identified factors independently associated with HIV-related chatbot use, including younger age, post-secondary education, being married, shorter ART duration, presence of comorbidities, smartphone ownership, internet access, and English proficiency.
This cross-sectional study surveyed 427 adults on antiretroviral treatment (ART) at a large tertiary referral center in Kano, Nigeria, finding that 75.2% were aware of AI chatbots, 72.6% had ever used one, and 66.5% had used chatbots for HIV-related queries (most commonly general HIV/ART information 36.3%, checking ART side effects 23.4%, and preparing questions for clinicians 15.0%), and identified factors independently associated with HIV-related chatbot use, including younger age, post-secondary education, being married, shorter ART duration, presence of comorbidities, smartphone ownership, internet access, and English proficiency.
Neurourology and Urodynamics This study developed and validated the Urodynamic Report Writer App (URapp), a GPT-4o-based virtual assistant modeled on scientific literature and international urodynamic guidelines, fine-tuned under supervision by two expert urodynamicists using examinations from multiple referral centers, and evaluated by functional urology specialists on a 0-5 Likert scale, scoring high on interpretative accuracy (4.5 ± 0.2), explanation of urodynamic concepts (5.0 ± 0), adherence to evidence-based standards (4.8 ± 0.3), clarity and practicality of recommendations (4.2 ± 0.3), clinical relevance (3.8 ± 0.3), and diagnostic usefulness (3.5 ± 0.5), with all 100 consecutive urodynamic studies (median age 61 yr) receiving overall scores ≥4.
This study developed and validated the Urodynamic Report Writer App (URapp), a GPT-4o-based virtual assistant modeled on scientific literature and international urodynamic guidelines, fine-tuned under supervision by two expert urodynamicists using examinations from multiple referral centers, and evaluated by functional urology specialists on a 0-5 Likert scale, scoring high on interpretative accuracy (4.5 ± 0.2), explanation of urodynamic concepts (5.0 ± 0), adherence to evidence-based standards (4.8 ± 0.3), clarity and practicality of recommendations (4.2 ± 0.3), clinical relevance (3.8 ± 0.3), and diagnostic usefulness (3.5 ± 0.5), with all 100 consecutive urodynamic studies (median age 61 yr) receiving overall scores ≥4.
This study developed and validated the Urodynamic Report Writer App (URapp), a GPT-4o-based virtual assistant modeled on scientific literature and international urodynamic guidelines, fine-tuned under supervision by two expert urodynamicists using examinations from multiple referral centers, and evaluated by functional urology specialists on a 0-5 Likert scale, scoring high on interpretative accuracy (4.5 ± 0.2), explanation of urodynamic concepts (5.0 ± 0), adherence to evidence-based standards (4.8 ± 0.3), clarity and practicality of recommendations (4.2 ± 0.3), clinical relevance (3.8 ± 0.3), and diagnostic usefulness (3.5 ± 0.5), with all 100 consecutive urodynamic studies (median age 61 yr) receiving overall scores ≥4.
This study developed and validated the Urodynamic Report Writer App (URapp), a GPT-4o-based virtual assistant modeled on scientific literature and international urodynamic guidelines, fine-tuned under supervision by two expert urodynamicists using examinations from multiple referral centers, and evaluated by functional urology specialists on a 0-5 Likert scale, scoring high on interpretative accuracy (4.5 ± 0.2), explanation of urodynamic concepts (5.0 ± 0), adherence to evidence-based standards (4.8 ± 0.3), clarity and practicality of recommendations (4.2 ± 0.3), clinical relevance (3.8 ± 0.3), and diagnostic usefulness (3.5 ± 0.5), with all 100 consecutive urodynamic studies (median age 61 yr) receiving overall scores ≥4.
Nature Medicine This Nature Medicine comment argues that trust in clinical AI cannot be benchmarked into existence but must be earned through rigorous prospective studies in real-world clinical settings, noting that the hardest lessons often concern the humans and systems around the AI rather than the technology itself.
This Nature Medicine comment argues that trust in clinical AI cannot be benchmarked into existence but must be earned through rigorous prospective studies in real-world clinical settings, noting that the hardest lessons often concern the humans and systems around the AI rather than the technology itself.
This Nature Medicine comment argues that trust in clinical AI cannot be benchmarked into existence but must be earned through rigorous prospective studies in real-world clinical settings, noting that the hardest lessons often concern the humans and systems around the AI rather than the technology itself.
This Nature Medicine comment argues that trust in clinical AI cannot be benchmarked into existence but must be earned through rigorous prospective studies in real-world clinical settings, noting that the hardest lessons often concern the humans and systems around the AI rather than the technology itself.
Journal of Medical Internet Research Using 5,467 Google Reviews posted in 2024 from all 24 medical centers in Taiwan, this study compared the common LDA dominant-topic assignment with GPT-based aspect-based sentiment analysis (ABSA) on the same corpus, finding that aspect-bearing reviews discussed an average of 2.05 distinct service aspects, that dominant-topic assignment yielded an illustrative 51.2% representational compression, that a soft-assignment LDA baseline reduced count-level compression to 1.7% but left semantic alignment limited (mean set Jaccard=0.33) and carried no aspect-level sentiment polarity, that 11.0% of multiaspect reviews showed cross-aspect mixed sentiment with Technical-Functional Divergence accounting for 61.
Using 5,467 Google Reviews posted in 2024 from all 24 medical centers in Taiwan, this study compared the common LDA dominant-topic assignment with GPT-based aspect-based sentiment analysis (ABSA) on the same corpus, finding that aspect-bearing reviews discussed an average of 2.05 distinct service aspects, that dominant-topic assignment yielded an illustrative 51.2% representational compression, that a soft-assignment LDA baseline reduced count-level compression to 1.7% but left semantic alignment limited (mean set Jaccard=0.33) and carried no aspect-level sentiment polarity, that 11.0% of multiaspect reviews showed cross-aspect mixed sentiment with Technical-Functional Divergence accounting for 61.
Using 5,467 Google Reviews posted in 2024 from all 24 medical centers in Taiwan, this study compared the common LDA dominant-topic assignment with GPT-based aspect-based sentiment analysis (ABSA) on the same corpus, finding that aspect-bearing reviews discussed an average of 2.05 distinct service aspects, that dominant-topic assignment yielded an illustrative 51.2% representational compression, that a soft-assignment LDA baseline reduced count-level compression to 1.7% but left semantic alignment limited (mean set Jaccard=0.33) and carried no aspect-level sentiment polarity, that 11.0% of multiaspect reviews showed cross-aspect mixed sentiment with Technical-Functional Divergence accounting for 61.
Using 5,467 Google Reviews posted in 2024 from all 24 medical centers in Taiwan, this study compared the common LDA dominant-topic assignment with GPT-based aspect-based sentiment analysis (ABSA) on the same corpus, finding that aspect-bearing reviews discussed an average of 2.05 distinct service aspects, that dominant-topic assignment yielded an illustrative 51.2% representational compression, that a soft-assignment LDA baseline reduced count-level compression to 1.7% but left semantic alignment limited (mean set Jaccard=0.33) and carried no aspect-level sentiment polarity, that 11.0% of multiaspect reviews showed cross-aspect mixed sentiment with Technical-Functional Divergence accounting for 61.
Nature A Nature news report describes a proof-of-concept study published in Nature Neuroscience in which a single surgically implanted 253-electrode array covering a fairly large area of the sensorimotor cortex, combined with artificial intelligence, simultaneously decoded attempted phrases and attempted or imagined gestures in two participants with impaired speech and movement after a brainstem stroke or with amyotrophic lateral sclerosis, producing on-screen text and driving a personalized animated avatar within seconds of the user’s intent, thereby translating both verbal and non-verbal communication through one implant.
A Nature news report describes a proof-of-concept study published in Nature Neuroscience in which a single surgically implanted 253-electrode array covering a fairly large area of the sensorimotor cortex, combined with artificial intelligence, simultaneously decoded attempted phrases and attempted or imagined gestures in two participants with impaired speech and movement after a brainstem stroke or with amyotrophic lateral sclerosis, producing on-screen text and driving a personalized animated avatar within seconds of the user’s intent, thereby translating both verbal and non-verbal communication through one implant.
A Nature news report describes a proof-of-concept study published in Nature Neuroscience in which a single surgically implanted 253-electrode array covering a fairly large area of the sensorimotor cortex, combined with artificial intelligence, simultaneously decoded attempted phrases and attempted or imagined gestures in two participants with impaired speech and movement after a brainstem stroke or with amyotrophic lateral sclerosis, producing on-screen text and driving a personalized animated avatar within seconds of the user’s intent, thereby translating both verbal and non-verbal communication through one implant.
A Nature news report describes a proof-of-concept study published in Nature Neuroscience in which a single surgically implanted 253-electrode array covering a fairly large area of the sensorimotor cortex, combined with artificial intelligence, simultaneously decoded attempted phrases and attempted or imagined gestures in two participants with impaired speech and movement after a brainstem stroke or with amyotrophic lateral sclerosis, producing on-screen text and driving a personalized animated avatar within seconds of the user’s intent, thereby translating both verbal and non-verbal communication through one implant.
Computational Materials Science This work introduces PolyGraphPy, an open-source unified Python framework that automates Density Functional Tight Binding (DFTB+) quantum-mechanics calculations to build structured datasets for monomers, homopolymers, and alternating copolymers, employs Bayesian Graph Neural Networks with stochastic graph representations for property prediction such as static polarizability together with uncertainty quantification, and incorporates two complementary generative models, a SELFIES-based Generative Pre-trained Transformer and a BRICS graph-fragmentation Genetic Algorithm, for de novo design of targeted molecules, demonstrated on a dataset of acrylates as a highly customizable end-to-end pipeline.
This work introduces PolyGraphPy, an open-source unified Python framework that automates Density Functional Tight Binding (DFTB+) quantum-mechanics calculations to build structured datasets for monomers, homopolymers, and alternating copolymers, employs Bayesian Graph Neural Networks with stochastic graph representations for property prediction such as static polarizability together with uncertainty quantification, and incorporates two complementary generative models, a SELFIES-based Generative Pre-trained Transformer and a BRICS graph-fragmentation Genetic Algorithm, for de novo design of targeted molecules, demonstrated on a dataset of acrylates as a highly customizable end-to-end pipeline.
This work introduces PolyGraphPy, an open-source unified Python framework that automates Density Functional Tight Binding (DFTB+) quantum-mechanics calculations to build structured datasets for monomers, homopolymers, and alternating copolymers, employs Bayesian Graph Neural Networks with stochastic graph representations for property prediction such as static polarizability together with uncertainty quantification, and incorporates two complementary generative models, a SELFIES-based Generative Pre-trained Transformer and a BRICS graph-fragmentation Genetic Algorithm, for de novo design of targeted molecules, demonstrated on a dataset of acrylates as a highly customizable end-to-end pipeline.
This work introduces PolyGraphPy, an open-source unified Python framework that automates Density Functional Tight Binding (DFTB+) quantum-mechanics calculations to build structured datasets for monomers, homopolymers, and alternating copolymers, employs Bayesian Graph Neural Networks with stochastic graph representations for property prediction such as static polarizability together with uncertainty quantification, and incorporates two complementary generative models, a SELFIES-based Generative Pre-trained Transformer and a BRICS graph-fragmentation Genetic Algorithm, for de novo design of targeted molecules, demonstrated on a dataset of acrylates as a highly customizable end-to-end pipeline.
Journal of Chemical Information and Modeling This work proposes SSE-DDI, a framework that performs selective substructure encoding in molecular line graphs with chemical bonds as the fundamental representation units, complemented by an edge-fusion graph transformer and refined SMILES-derived Morgan-fingerprint similarity profiles; on DrugBank and Twosides under transductive and inductive settings it outperforms representative baselines across multiple metrics, with ablation and visualization analyses supporting the effectiveness of selective encoding and highlighting DDI-relevant molecular substructures.
This work proposes SSE-DDI, a framework that performs selective substructure encoding in molecular line graphs with chemical bonds as the fundamental representation units, complemented by an edge-fusion graph transformer and refined SMILES-derived Morgan-fingerprint similarity profiles; on DrugBank and Twosides under transductive and inductive settings it outperforms representative baselines across multiple metrics, with ablation and visualization analyses supporting the effectiveness of selective encoding and highlighting DDI-relevant molecular substructures.
This work proposes SSE-DDI, a framework that performs selective substructure encoding in molecular line graphs with chemical bonds as the fundamental representation units, complemented by an edge-fusion graph transformer and refined SMILES-derived Morgan-fingerprint similarity profiles; on DrugBank and Twosides under transductive and inductive settings it outperforms representative baselines across multiple metrics, with ablation and visualization analyses supporting the effectiveness of selective encoding and highlighting DDI-relevant molecular substructures.
This work proposes SSE-DDI, a framework that performs selective substructure encoding in molecular line graphs with chemical bonds as the fundamental representation units, complemented by an edge-fusion graph transformer and refined SMILES-derived Morgan-fingerprint similarity profiles; on DrugBank and Twosides under transductive and inductive settings it outperforms representative baselines across multiple metrics, with ablation and visualization analyses supporting the effectiveness of selective encoding and highlighting DDI-relevant molecular substructures.
Practical Radiation Oncology This single-center prospective pilot study first implemented and evaluated a fully automatic All-in-One radiotherapy workflow on an integrated CT-linac in 20 patients with rectal cancer, sequentially completing CT simulation, AI-based autosegmentation, autoplanning, online verification, and beam delivery on one platform without patient repositioning; the workflow was completed in all patients, mean total treatment time was 37.5 ± 6.9 minutes, Dice similarity coefficients ranged from 0.83 to 0.95, three-dimensional gamma pass rates exceeded 95% in all patients, leukopenia occurred in 55.0%, and at follow-up 19 of 20 patients underwent surgery with 1 clinical complete response, an overall complete response rate of 15.0% and tumor regression grade 0-1 in 68.4% of surgical patients.
This single-center prospective pilot study first implemented and evaluated a fully automatic All-in-One radiotherapy workflow on an integrated CT-linac in 20 patients with rectal cancer, sequentially completing CT simulation, AI-based autosegmentation, autoplanning, online verification, and beam delivery on one platform without patient repositioning; the workflow was completed in all patients, mean total treatment time was 37.5 ± 6.9 minutes, Dice similarity coefficients ranged from 0.83 to 0.95, three-dimensional gamma pass rates exceeded 95% in all patients, leukopenia occurred in 55.0%, and at follow-up 19 of 20 patients underwent surgery with 1 clinical complete response, an overall complete response rate of 15.0% and tumor regression grade 0-1 in 68.4% of surgical patients.
This single-center prospective pilot study first implemented and evaluated a fully automatic All-in-One radiotherapy workflow on an integrated CT-linac in 20 patients with rectal cancer, sequentially completing CT simulation, AI-based autosegmentation, autoplanning, online verification, and beam delivery on one platform without patient repositioning; the workflow was completed in all patients, mean total treatment time was 37.5 ± 6.9 minutes, Dice similarity coefficients ranged from 0.83 to 0.95, three-dimensional gamma pass rates exceeded 95% in all patients, leukopenia occurred in 55.0%, and at follow-up 19 of 20 patients underwent surgery with 1 clinical complete response, an overall complete response rate of 15.0% and tumor regression grade 0-1 in 68.4% of surgical patients.
This single-center prospective pilot study first implemented and evaluated a fully automatic All-in-One radiotherapy workflow on an integrated CT-linac in 20 patients with rectal cancer, sequentially completing CT simulation, AI-based autosegmentation, autoplanning, online verification, and beam delivery on one platform without patient repositioning; the workflow was completed in all patients, mean total treatment time was 37.5 ± 6.9 minutes, Dice similarity coefficients ranged from 0.83 to 0.95, three-dimensional gamma pass rates exceeded 95% in all patients, leukopenia occurred in 55.0%, and at follow-up 19 of 20 patients underwent surgery with 1 clinical complete response, an overall complete response rate of 15.0% and tumor regression grade 0-1 in 68.4% of surgical patients.
Current Neurology and Neuroscience Reports This review evaluates the current state of multimodal monitoring (MMM) technologies in traumatic brain injury (TBI), their clinical applications, and their emerging role in improving TBI management, focusing on real-time physiological data streams such as intracranial pressure, brain tissue oxygenation, arterial blood pressure, and electroencephalography; it suggests that integrating physiological variables with artificial intelligence and advanced analytics may enable earlier detection and prediction of adverse events such as seizures, brain hypoxic events, and excursions of cerebral hypertension, potentially reducing variability in TBI outcomes associated with secondary insults, while the optimal combination of monitoring modalities, interpretation of complex data streams, larger harmoni
This review evaluates the current state of multimodal monitoring (MMM) technologies in traumatic brain injury (TBI), their clinical applications, and their emerging role in improving TBI management, focusing on real-time physiological data streams such as intracranial pressure, brain tissue oxygenation, arterial blood pressure, and electroencephalography; it suggests that integrating physiological variables with artificial intelligence and advanced analytics may enable earlier detection and prediction of adverse events such as seizures, brain hypoxic events, and excursions of cerebral hypertension, potentially reducing variability in TBI outcomes associated with secondary insults, while the optimal combination of monitoring modalities, interpretation of complex data streams, larger harmoni
This review evaluates the current state of multimodal monitoring (MMM) technologies in traumatic brain injury (TBI), their clinical applications, and their emerging role in improving TBI management, focusing on real-time physiological data streams such as intracranial pressure, brain tissue oxygenation, arterial blood pressure, and electroencephalography; it suggests that integrating physiological variables with artificial intelligence and advanced analytics may enable earlier detection and prediction of adverse events such as seizures, brain hypoxic events, and excursions of cerebral hypertension, potentially reducing variability in TBI outcomes associated with secondary insults, while the optimal combination of monitoring modalities, interpretation of complex data streams, larger harmoni
This review evaluates the current state of multimodal monitoring (MMM) technologies in traumatic brain injury (TBI), their clinical applications, and their emerging role in improving TBI management, focusing on real-time physiological data streams such as intracranial pressure, brain tissue oxygenation, arterial blood pressure, and electroencephalography; it suggests that integrating physiological variables with artificial intelligence and advanced analytics may enable earlier detection and prediction of adverse events such as seizures, brain hypoxic events, and excursions of cerebral hypertension, potentially reducing variability in TBI outcomes associated with secondary insults, while the optimal combination of monitoring modalities, interpretation of complex data streams, larger harmoni
medRxiv This study conducted a systematic mapping and bibliometric analysis of 4,143 publications from 2015 to 2025 on AI-generated synthetic data in biomedicine, combining expert annotation with LLM-assisted classification across data modality, medical domain, paper type, deployment status, and research stance, finding continuous growth in publication volume, 77.8% of papers strongly supportive with critical work below 1%, medical imaging dominating the corpus, highly cited primary research concentrated in molecular and pharmaceutical applications, and only 27 publications reporting operational use, thereby revealing a gap between methodological growth and deployment.
This study conducted a systematic mapping and bibliometric analysis of 4,143 publications from 2015 to 2025 on AI-generated synthetic data in biomedicine, combining expert annotation with LLM-assisted classification across data modality, medical domain, paper type, deployment status, and research stance, finding continuous growth in publication volume, 77.8% of papers strongly supportive with critical work below 1%, medical imaging dominating the corpus, highly cited primary research concentrated in molecular and pharmaceutical applications, and only 27 publications reporting operational use, thereby revealing a gap between methodological growth and deployment.
This study conducted a systematic mapping and bibliometric analysis of 4,143 publications from 2015 to 2025 on AI-generated synthetic data in biomedicine, combining expert annotation with LLM-assisted classification across data modality, medical domain, paper type, deployment status, and research stance, finding continuous growth in publication volume, 77.8% of papers strongly supportive with critical work below 1%, medical imaging dominating the corpus, highly cited primary research concentrated in molecular and pharmaceutical applications, and only 27 publications reporting operational use, thereby revealing a gap between methodological growth and deployment.
This study conducted a systematic mapping and bibliometric analysis of 4,143 publications from 2015 to 2025 on AI-generated synthetic data in biomedicine, combining expert annotation with LLM-assisted classification across data modality, medical domain, paper type, deployment status, and research stance, finding continuous growth in publication volume, 77.8% of papers strongly supportive with critical work below 1%, medical imaging dominating the corpus, highly cited primary research concentrated in molecular and pharmaceutical applications, and only 27 publications reporting operational use, thereby revealing a gap between methodological growth and deployment.
arXiv This work proposes Top-Theta (Top-j) Attention, a training-free inference-time method that replaces top-k search with static per-head calibrated thresholds to sparsify attention elements, and introduces Softmax Denominator Compensation (SDC) and V-Mean Compensation (VMC) to preserve accuracy, achieving 3–10× reduction in V-cache rows and up to 10× fewer attention elements on LLaMA2/LLaMA3 Q&A, code generation, and long-context summarization tasks with no more than 1% accuracy degradation.
This work proposes Top-Theta (Top-j) Attention, a training-free inference-time method that replaces top-k search with static per-head calibrated thresholds to sparsify attention elements, and introduces Softmax Denominator Compensation (SDC) and V-Mean Compensation (VMC) to preserve accuracy, achieving 3–10× reduction in V-cache rows and up to 10× fewer attention elements on LLaMA2/LLaMA3 Q&A, code generation, and long-context summarization tasks with no more than 1% accuracy degradation.
This work proposes Top-Theta (Top-j) Attention, a training-free inference-time method that replaces top-k search with static per-head calibrated thresholds to sparsify attention elements, and introduces Softmax Denominator Compensation (SDC) and V-Mean Compensation (VMC) to preserve accuracy, achieving 3–10× reduction in V-cache rows and up to 10× fewer attention elements on LLaMA2/LLaMA3 Q&A, code generation, and long-context summarization tasks with no more than 1% accuracy degradation.
This work proposes Top-Theta (Top-j) Attention, a training-free inference-time method that replaces top-k search with static per-head calibrated thresholds to sparsify attention elements, and introduces Softmax Denominator Compensation (SDC) and V-Mean Compensation (VMC) to preserve accuracy, achieving 3–10× reduction in V-cache rows and up to 10× fewer attention elements on LLaMA2/LLaMA3 Q&A, code generation, and long-context summarization tasks with no more than 1% accuracy degradation.
Drug Discovery Today The article proposes a framework for natural product drug discovery that translates evidence into auditable experimental actions through a four-tier architecture (source authentication, reproducible chemical fingerprinting, orthogonal prioritization, definitive characterization) with documented decision gates, and an 'evidence ladder' that separates molecular-identification confidence from chemical novelty, biological novelty and translational relevance, addressing three recurring gaps: incomplete integration of sample metadata, conflation of analytical detection with molecular novelty, and lack of explicit decision thresholds.
The article proposes a framework for natural product drug discovery that translates evidence into auditable experimental actions through a four-tier architecture (source authentication, reproducible chemical fingerprinting, orthogonal prioritization, definitive characterization) with documented decision gates, and an 'evidence ladder' that separates molecular-identification confidence from chemical novelty, biological novelty and translational relevance, addressing three recurring gaps: incomplete integration of sample metadata, conflation of analytical detection with molecular novelty, and lack of explicit decision thresholds.
The article proposes a framework for natural product drug discovery that translates evidence into auditable experimental actions through a four-tier architecture (source authentication, reproducible chemical fingerprinting, orthogonal prioritization, definitive characterization) with documented decision gates, and an 'evidence ladder' that separates molecular-identification confidence from chemical novelty, biological novelty and translational relevance, addressing three recurring gaps: incomplete integration of sample metadata, conflation of analytical detection with molecular novelty, and lack of explicit decision thresholds.
The article proposes a framework for natural product drug discovery that translates evidence into auditable experimental actions through a four-tier architecture (source authentication, reproducible chemical fingerprinting, orthogonal prioritization, definitive characterization) with documented decision gates, and an 'evidence ladder' that separates molecular-identification confidence from chemical novelty, biological novelty and translational relevance, addressing three recurring gaps: incomplete integration of sample metadata, conflation of analytical detection with molecular novelty, and lack of explicit decision thresholds.
Journal of Chemical Information and Modeling This study introduces a new avenue for foundation model pretraining: supervised pretraining on low-noise, calculable molecular descriptors to obtain rich, highly transferable molecular representations, demonstrated with CheMeleon, an O(10M) parameter foundation model; across 58 benchmark data sets spanning properties relevant to small-molecule drug discovery and sourced from the industry-led Polaris benchmarking initiative, CheMeleon enables directed message-passing neural networks to finally exceed classical methods in the low-data regime, outperforms classical baselines such as Random Forest on molecular fingerprints and descriptors as well as existing foundation models under rigorous statistical comparisons, and the model and pretraining framework are open-sourced.
This study introduces a new avenue for foundation model pretraining: supervised pretraining on low-noise, calculable molecular descriptors to obtain rich, highly transferable molecular representations, demonstrated with CheMeleon, an O(10M) parameter foundation model; across 58 benchmark data sets spanning properties relevant to small-molecule drug discovery and sourced from the industry-led Polaris benchmarking initiative, CheMeleon enables directed message-passing neural networks to finally exceed classical methods in the low-data regime, outperforms classical baselines such as Random Forest on molecular fingerprints and descriptors as well as existing foundation models under rigorous statistical comparisons, and the model and pretraining framework are open-sourced.
This study introduces a new avenue for foundation model pretraining: supervised pretraining on low-noise, calculable molecular descriptors to obtain rich, highly transferable molecular representations, demonstrated with CheMeleon, an O(10M) parameter foundation model; across 58 benchmark data sets spanning properties relevant to small-molecule drug discovery and sourced from the industry-led Polaris benchmarking initiative, CheMeleon enables directed message-passing neural networks to finally exceed classical methods in the low-data regime, outperforms classical baselines such as Random Forest on molecular fingerprints and descriptors as well as existing foundation models under rigorous statistical comparisons, and the model and pretraining framework are open-sourced.
This study introduces a new avenue for foundation model pretraining: supervised pretraining on low-noise, calculable molecular descriptors to obtain rich, highly transferable molecular representations, demonstrated with CheMeleon, an O(10M) parameter foundation model; across 58 benchmark data sets spanning properties relevant to small-molecule drug discovery and sourced from the industry-led Polaris benchmarking initiative, CheMeleon enables directed message-passing neural networks to finally exceed classical methods in the low-data regime, outperforms classical baselines such as Random Forest on molecular fingerprints and descriptors as well as existing foundation models under rigorous statistical comparisons, and the model and pretraining framework are open-sourced.
Intelligent Computing This survey systematically reviews prompt engineering research for SAM and its growing ecosystem, proposing a hierarchical taxonomy that organizes methods into geometric prompts, textual semantic prompts, and multimodal fusion prompts, further tracing the transition from manually crafted prompts to automated generation based on detector outputs, prototype learning, reinforcement learning, and vision-language models, while tracing how prompt engineering enables cross-domain generalization in medical imaging, remote sensing, industrial inspection, and anomaly detection, and identifying key challenges such as prompt sensitivity, cross-modal misalignment, and computational inefficiency alongside future directions including causal prompt reasoning, collaborative multi-agent prompting, and diffu
This survey systematically reviews prompt engineering research for SAM and its growing ecosystem, proposing a hierarchical taxonomy that organizes methods into geometric prompts, textual semantic prompts, and multimodal fusion prompts, further tracing the transition from manually crafted prompts to automated generation based on detector outputs, prototype learning, reinforcement learning, and vision-language models, while tracing how prompt engineering enables cross-domain generalization in medical imaging, remote sensing, industrial inspection, and anomaly detection, and identifying key challenges such as prompt sensitivity, cross-modal misalignment, and computational inefficiency alongside future directions including causal prompt reasoning, collaborative multi-agent prompting, and diffu
This survey systematically reviews prompt engineering research for SAM and its growing ecosystem, proposing a hierarchical taxonomy that organizes methods into geometric prompts, textual semantic prompts, and multimodal fusion prompts, further tracing the transition from manually crafted prompts to automated generation based on detector outputs, prototype learning, reinforcement learning, and vision-language models, while tracing how prompt engineering enables cross-domain generalization in medical imaging, remote sensing, industrial inspection, and anomaly detection, and identifying key challenges such as prompt sensitivity, cross-modal misalignment, and computational inefficiency alongside future directions including causal prompt reasoning, collaborative multi-agent prompting, and diffu
This survey systematically reviews prompt engineering research for SAM and its growing ecosystem, proposing a hierarchical taxonomy that organizes methods into geometric prompts, textual semantic prompts, and multimodal fusion prompts, further tracing the transition from manually crafted prompts to automated generation based on detector outputs, prototype learning, reinforcement learning, and vision-language models, while tracing how prompt engineering enables cross-domain generalization in medical imaging, remote sensing, industrial inspection, and anomaly detection, and identifying key challenges such as prompt sensitivity, cross-modal misalignment, and computational inefficiency alongside future directions including causal prompt reasoning, collaborative multi-agent prompting, and diffu
JMIR Medical Informatics This study deployed the open-source Qwen3-14B model on a hospital intranet server using the Ollama framework, supported it with lightweight knowledge augmentation through exact-match injection from a structured knowledge base derived from drug package inserts, and used a 2-period crossover design in which 2 pharmacists independently reviewed the same 213 outpatient prescriptions under unaided and AI-assisted conditions; human-AI collaborative review achieved 97.2% (207/213) accuracy versus 82.6% (176/213) for pharmacist-alone review, sensitivity was 98% (61/62) versus 55% (34/62), the false-negative rate fell from 45% to 2%, knowledge augmentation reduced the model hallucination rate from 19.7% (42/213) to 4.7% (10/213), mean per-prescription review time fell from 2.33 to 1.
This study deployed the open-source Qwen3-14B model on a hospital intranet server using the Ollama framework, supported it with lightweight knowledge augmentation through exact-match injection from a structured knowledge base derived from drug package inserts, and used a 2-period crossover design in which 2 pharmacists independently reviewed the same 213 outpatient prescriptions under unaided and AI-assisted conditions; human-AI collaborative review achieved 97.2% (207/213) accuracy versus 82.6% (176/213) for pharmacist-alone review, sensitivity was 98% (61/62) versus 55% (34/62), the false-negative rate fell from 45% to 2%, knowledge augmentation reduced the model hallucination rate from 19.7% (42/213) to 4.7% (10/213), mean per-prescription review time fell from 2.33 to 1.
This study deployed the open-source Qwen3-14B model on a hospital intranet server using the Ollama framework, supported it with lightweight knowledge augmentation through exact-match injection from a structured knowledge base derived from drug package inserts, and used a 2-period crossover design in which 2 pharmacists independently reviewed the same 213 outpatient prescriptions under unaided and AI-assisted conditions; human-AI collaborative review achieved 97.2% (207/213) accuracy versus 82.6% (176/213) for pharmacist-alone review, sensitivity was 98% (61/62) versus 55% (34/62), the false-negative rate fell from 45% to 2%, knowledge augmentation reduced the model hallucination rate from 19.7% (42/213) to 4.7% (10/213), mean per-prescription review time fell from 2.33 to 1.
This study deployed the open-source Qwen3-14B model on a hospital intranet server using the Ollama framework, supported it with lightweight knowledge augmentation through exact-match injection from a structured knowledge base derived from drug package inserts, and used a 2-period crossover design in which 2 pharmacists independently reviewed the same 213 outpatient prescriptions under unaided and AI-assisted conditions; human-AI collaborative review achieved 97.2% (207/213) accuracy versus 82.6% (176/213) for pharmacist-alone review, sensitivity was 98% (61/62) versus 55% (34/62), the false-negative rate fell from 45% to 2%, knowledge augmentation reduced the model hallucination rate from 19.7% (42/213) to 4.7% (10/213), mean per-prescription review time fell from 2.33 to 1.
New Biotechnology This study treats mRNA stability as a tunable codon-optimization design parameter: it built a combinatorial library of synonymous coding-sequence variants of an IgG1 light chain integrated as single copies at a defined genomic locus in CHO cells, used steady-state mRNA abundance quantified by deep sequencing of gDNA and mRNA as a proxy for stability, trained a machine-learning model predicting mRNA abundance from coding sequence using embeddings from a pre-trained nucleotide transformer, and incorporated this predictor with established translational metrics into a genetic algorithm for multi-objective codon optimization; as proof-of-concept with Trastuzumab-encoding sequences, high-abundance designs raised intracellular mRNA by 41%, protein titer by 59%, and cell-specific productivity by 8
This study treats mRNA stability as a tunable codon-optimization design parameter: it built a combinatorial library of synonymous coding-sequence variants of an IgG1 light chain integrated as single copies at a defined genomic locus in CHO cells, used steady-state mRNA abundance quantified by deep sequencing of gDNA and mRNA as a proxy for stability, trained a machine-learning model predicting mRNA abundance from coding sequence using embeddings from a pre-trained nucleotide transformer, and incorporated this predictor with established translational metrics into a genetic algorithm for multi-objective codon optimization; as proof-of-concept with Trastuzumab-encoding sequences, high-abundance designs raised intracellular mRNA by 41%, protein titer by 59%, and cell-specific productivity by 8
This study treats mRNA stability as a tunable codon-optimization design parameter: it built a combinatorial library of synonymous coding-sequence variants of an IgG1 light chain integrated as single copies at a defined genomic locus in CHO cells, used steady-state mRNA abundance quantified by deep sequencing of gDNA and mRNA as a proxy for stability, trained a machine-learning model predicting mRNA abundance from coding sequence using embeddings from a pre-trained nucleotide transformer, and incorporated this predictor with established translational metrics into a genetic algorithm for multi-objective codon optimization; as proof-of-concept with Trastuzumab-encoding sequences, high-abundance designs raised intracellular mRNA by 41%, protein titer by 59%, and cell-specific productivity by 8
This study treats mRNA stability as a tunable codon-optimization design parameter: it built a combinatorial library of synonymous coding-sequence variants of an IgG1 light chain integrated as single copies at a defined genomic locus in CHO cells, used steady-state mRNA abundance quantified by deep sequencing of gDNA and mRNA as a proxy for stability, trained a machine-learning model predicting mRNA abundance from coding sequence using embeddings from a pre-trained nucleotide transformer, and incorporated this predictor with established translational metrics into a genetic algorithm for multi-objective codon optimization; as proof-of-concept with Trastuzumab-encoding sequences, high-abundance designs raised intracellular mRNA by 41%, protein titer by 59%, and cell-specific productivity by 8
Neurosurgical Review This systematic review and meta-analysis pooled 5 studies covering 577 patients with confirmed trigeminal neuralgia to evaluate deep learning and machine learning models for automated trigeminal nerve segmentation and neurovascular conflict detection on MRI, finding a pooled sensitivity of 71% (95% CI: 65-76%), specificity of 92% (95% CI: 88-94%), and AUC of 0.91 (95% CI: 0.88-0.93), suggesting high diagnostic accuracy for AI models in this task.
This systematic review and meta-analysis pooled 5 studies covering 577 patients with confirmed trigeminal neuralgia to evaluate deep learning and machine learning models for automated trigeminal nerve segmentation and neurovascular conflict detection on MRI, finding a pooled sensitivity of 71% (95% CI: 65-76%), specificity of 92% (95% CI: 88-94%), and AUC of 0.91 (95% CI: 0.88-0.93), suggesting high diagnostic accuracy for AI models in this task.
This systematic review and meta-analysis pooled 5 studies covering 577 patients with confirmed trigeminal neuralgia to evaluate deep learning and machine learning models for automated trigeminal nerve segmentation and neurovascular conflict detection on MRI, finding a pooled sensitivity of 71% (95% CI: 65-76%), specificity of 92% (95% CI: 88-94%), and AUC of 0.91 (95% CI: 0.88-0.93), suggesting high diagnostic accuracy for AI models in this task.
This systematic review and meta-analysis pooled 5 studies covering 577 patients with confirmed trigeminal neuralgia to evaluate deep learning and machine learning models for automated trigeminal nerve segmentation and neurovascular conflict detection on MRI, finding a pooled sensitivity of 71% (95% CI: 65-76%), specificity of 92% (95% CI: 88-94%), and AUC of 0.91 (95% CI: 0.88-0.93), suggesting high diagnostic accuracy for AI models in this task.
arXiv Using vacancies in the semiconductor Sb2Se3 as a case study, this work finds that current foundation machine learning interatomic potentials trained on bulk data do not reliably identify defect ground states, and introduces global defect charge embeddings in the MACE architecture together with a multi-fidelity training strategy combining low-cost PBE data with high-quality HSE06 data, achieving ground-state identification within 0.05 Å and defect formation energies and thermodynamic transition levels within 0.02 eV of hybrid-functional DFT at a fixed defect supercell size, while uncovering global minima missed by standard defect search workflows.
Using vacancies in the semiconductor Sb2Se3 as a case study, this work finds that current foundation machine learning interatomic potentials trained on bulk data do not reliably identify defect ground states, and introduces global defect charge embeddings in the MACE architecture together with a multi-fidelity training strategy combining low-cost PBE data with high-quality HSE06 data, achieving ground-state identification within 0.05 Å and defect formation energies and thermodynamic transition levels within 0.02 eV of hybrid-functional DFT at a fixed defect supercell size, while uncovering global minima missed by standard defect search workflows.
Using vacancies in the semiconductor Sb2Se3 as a case study, this work finds that current foundation machine learning interatomic potentials trained on bulk data do not reliably identify defect ground states, and introduces global defect charge embeddings in the MACE architecture together with a multi-fidelity training strategy combining low-cost PBE data with high-quality HSE06 data, achieving ground-state identification within 0.05 Å and defect formation energies and thermodynamic transition levels within 0.02 eV of hybrid-functional DFT at a fixed defect supercell size, while uncovering global minima missed by standard defect search workflows.
Using vacancies in the semiconductor Sb2Se3 as a case study, this work finds that current foundation machine learning interatomic potentials trained on bulk data do not reliably identify defect ground states, and introduces global defect charge embeddings in the MACE architecture together with a multi-fidelity training strategy combining low-cost PBE data with high-quality HSE06 data, achieving ground-state identification within 0.05 Å and defect formation energies and thermodynamic transition levels within 0.02 eV of hybrid-functional DFT at a fixed defect supercell size, while uncovering global minima missed by standard defect search workflows.
Biomedical Microdevices This review examines how microwell platforms determine the information obtainable from individual cells through cell-loading strategies, well geometry, platform architecture and material selection, reviews their applications in cellular behaviour and cell-cell interactions, secretome analysis, genomic and transcriptomic profiling, and drug screening and precision medicine, and distinguishes AI approaches directly demonstrated in microwell-based studies from those still prospective, indicating that linking microwell engineering with AI-driven analysis can move microwell-based single-cell research from measurement towards prediction and autonomous discovery.
This review examines how microwell platforms determine the information obtainable from individual cells through cell-loading strategies, well geometry, platform architecture and material selection, reviews their applications in cellular behaviour and cell-cell interactions, secretome analysis, genomic and transcriptomic profiling, and drug screening and precision medicine, and distinguishes AI approaches directly demonstrated in microwell-based studies from those still prospective, indicating that linking microwell engineering with AI-driven analysis can move microwell-based single-cell research from measurement towards prediction and autonomous discovery.
This review examines how microwell platforms determine the information obtainable from individual cells through cell-loading strategies, well geometry, platform architecture and material selection, reviews their applications in cellular behaviour and cell-cell interactions, secretome analysis, genomic and transcriptomic profiling, and drug screening and precision medicine, and distinguishes AI approaches directly demonstrated in microwell-based studies from those still prospective, indicating that linking microwell engineering with AI-driven analysis can move microwell-based single-cell research from measurement towards prediction and autonomous discovery.
This review examines how microwell platforms determine the information obtainable from individual cells through cell-loading strategies, well geometry, platform architecture and material selection, reviews their applications in cellular behaviour and cell-cell interactions, secretome analysis, genomic and transcriptomic profiling, and drug screening and precision medicine, and distinguishes AI approaches directly demonstrated in microwell-based studies from those still prospective, indicating that linking microwell engineering with AI-driven analysis can move microwell-based single-cell research from measurement towards prediction and autonomous discovery.
Magnetic resonance in medicine This work proposes ESPIRiT-Diffusion, a physics-guided score-based diffusion reconstruction framework that incorporates multi-set ESPIRiT coil sensitivity map-based data-consistency constraints into the Langevin equation for 8.8- and 10.7-fold accelerated joint intracranial and carotid vessel wall imaging (VWI) at 0.6 mm³ isotropic resolution; in retrospective experiments with Cartesian, CAIPI, and variable-density undersampling it showed improved reconstruction performance compared with ESPIRiT, DL-ESPIRiT, SENSE-Diffusion, and SPIRiT-Diffusion with better preservation of fine vessel wall structures, and in prospective patient experiments it provided favorable visualization of vessel wall lesions with no statistically significant differences in reader scores from the 3-fold CS reference a
This work proposes ESPIRiT-Diffusion, a physics-guided score-based diffusion reconstruction framework that incorporates multi-set ESPIRiT coil sensitivity map-based data-consistency constraints into the Langevin equation for 8.8- and 10.7-fold accelerated joint intracranial and carotid vessel wall imaging (VWI) at 0.6 mm³ isotropic resolution; in retrospective experiments with Cartesian, CAIPI, and variable-density undersampling it showed improved reconstruction performance compared with ESPIRiT, DL-ESPIRiT, SENSE-Diffusion, and SPIRiT-Diffusion with better preservation of fine vessel wall structures, and in prospective patient experiments it provided favorable visualization of vessel wall lesions with no statistically significant differences in reader scores from the 3-fold CS reference a
This work proposes ESPIRiT-Diffusion, a physics-guided score-based diffusion reconstruction framework that incorporates multi-set ESPIRiT coil sensitivity map-based data-consistency constraints into the Langevin equation for 8.8- and 10.7-fold accelerated joint intracranial and carotid vessel wall imaging (VWI) at 0.6 mm³ isotropic resolution; in retrospective experiments with Cartesian, CAIPI, and variable-density undersampling it showed improved reconstruction performance compared with ESPIRiT, DL-ESPIRiT, SENSE-Diffusion, and SPIRiT-Diffusion with better preservation of fine vessel wall structures, and in prospective patient experiments it provided favorable visualization of vessel wall lesions with no statistically significant differences in reader scores from the 3-fold CS reference a
This work proposes ESPIRiT-Diffusion, a physics-guided score-based diffusion reconstruction framework that incorporates multi-set ESPIRiT coil sensitivity map-based data-consistency constraints into the Langevin equation for 8.8- and 10.7-fold accelerated joint intracranial and carotid vessel wall imaging (VWI) at 0.6 mm³ isotropic resolution; in retrospective experiments with Cartesian, CAIPI, and variable-density undersampling it showed improved reconstruction performance compared with ESPIRiT, DL-ESPIRiT, SENSE-Diffusion, and SPIRiT-Diffusion with better preservation of fine vessel wall structures, and in prospective patient experiments it provided favorable visualization of vessel wall lesions with no statistically significant differences in reader scores from the 3-fold CS reference a
Facial Plastic Surgery This is a forward-looking article examining the emerging role of personalized medicine in facial plastic surgery, proposing that biologically, anatomically, and psychologically tailored approaches may refine both aesthetic and reconstructive care, and suggesting that genomics, pharmacogenomics, artificial intelligence, tissue engineering, and three-dimensional modelling may improve prediction of healing, treatment response, complication risk, and reconstructive requirements, while emphasizing that ethical challenges relating to privacy, bias, and equitable access must remain central.
This is a forward-looking article examining the emerging role of personalized medicine in facial plastic surgery, proposing that biologically, anatomically, and psychologically tailored approaches may refine both aesthetic and reconstructive care, and suggesting that genomics, pharmacogenomics, artificial intelligence, tissue engineering, and three-dimensional modelling may improve prediction of healing, treatment response, complication risk, and reconstructive requirements, while emphasizing that ethical challenges relating to privacy, bias, and equitable access must remain central.
This is a forward-looking article examining the emerging role of personalized medicine in facial plastic surgery, proposing that biologically, anatomically, and psychologically tailored approaches may refine both aesthetic and reconstructive care, and suggesting that genomics, pharmacogenomics, artificial intelligence, tissue engineering, and three-dimensional modelling may improve prediction of healing, treatment response, complication risk, and reconstructive requirements, while emphasizing that ethical challenges relating to privacy, bias, and equitable access must remain central.
This is a forward-looking article examining the emerging role of personalized medicine in facial plastic surgery, proposing that biologically, anatomically, and psychologically tailored approaches may refine both aesthetic and reconstructive care, and suggesting that genomics, pharmacogenomics, artificial intelligence, tissue engineering, and three-dimensional modelling may improve prediction of healing, treatment response, complication risk, and reconstructive requirements, while emphasizing that ethical challenges relating to privacy, bias, and equitable access must remain central.
arXiv The work proposes a Hierarchical Diffusion Classifier (HDC) that exploits parent-child label hierarchies: in a pruning stage it traverses the label tree level by level, estimates epsilon-prediction errors with fewer Monte Carlo samples, and keeps only the lowest-error nodes, then runs classical diffusion classification on the surviving leaf nodes, achieving roughly 60% inference speed-up on ImageNet-1K with Stable Diffusion 2.0 (1600s to 650s) or raising per-class accuracy from 64.90% to 65.16% at nearly unchanged runtime.
The work proposes a Hierarchical Diffusion Classifier (HDC) that exploits parent-child label hierarchies: in a pruning stage it traverses the label tree level by level, estimates epsilon-prediction errors with fewer Monte Carlo samples, and keeps only the lowest-error nodes, then runs classical diffusion classification on the surviving leaf nodes, achieving roughly 60% inference speed-up on ImageNet-1K with Stable Diffusion 2.0 (1600s to 650s) or raising per-class accuracy from 64.90% to 65.16% at nearly unchanged runtime.
The work proposes a Hierarchical Diffusion Classifier (HDC) that exploits parent-child label hierarchies: in a pruning stage it traverses the label tree level by level, estimates epsilon-prediction errors with fewer Monte Carlo samples, and keeps only the lowest-error nodes, then runs classical diffusion classification on the surviving leaf nodes, achieving roughly 60% inference speed-up on ImageNet-1K with Stable Diffusion 2.0 (1600s to 650s) or raising per-class accuracy from 64.90% to 65.16% at nearly unchanged runtime.
The work proposes a Hierarchical Diffusion Classifier (HDC) that exploits parent-child label hierarchies: in a pruning stage it traverses the label tree level by level, estimates epsilon-prediction errors with fewer Monte Carlo samples, and keeps only the lowest-error nodes, then runs classical diffusion classification on the surviving leaf nodes, achieving roughly 60% inference speed-up on ImageNet-1K with Stable Diffusion 2.0 (1600s to 650s) or raising per-class accuracy from 64.90% to 65.16% at nearly unchanged runtime.
medRxiv This study developed CARDIAC-FM, a multimodal foundation model that integrates 12-lead ECG and cardiac MRI through self-supervised representation learning and cross-modal contrastive alignment, pretrained on 57,609 paired samples from UK Biobank, improving prediction of incident atrial fibrillation and heart failure, generalizing zero-shot to the Cardiovascular Health Study and the Multi-Ethnic Study of Atherosclerosis, and transferring to cardiac MRI phenotype prediction, time-to-event modelling, and additional outcomes including myocardial infarction, ischaemic stroke, cardiovascular death, and all-cause mortality.
This study developed CARDIAC-FM, a multimodal foundation model that integrates 12-lead ECG and cardiac MRI through self-supervised representation learning and cross-modal contrastive alignment, pretrained on 57,609 paired samples from UK Biobank, improving prediction of incident atrial fibrillation and heart failure, generalizing zero-shot to the Cardiovascular Health Study and the Multi-Ethnic Study of Atherosclerosis, and transferring to cardiac MRI phenotype prediction, time-to-event modelling, and additional outcomes including myocardial infarction, ischaemic stroke, cardiovascular death, and all-cause mortality.
This study developed CARDIAC-FM, a multimodal foundation model that integrates 12-lead ECG and cardiac MRI through self-supervised representation learning and cross-modal contrastive alignment, pretrained on 57,609 paired samples from UK Biobank, improving prediction of incident atrial fibrillation and heart failure, generalizing zero-shot to the Cardiovascular Health Study and the Multi-Ethnic Study of Atherosclerosis, and transferring to cardiac MRI phenotype prediction, time-to-event modelling, and additional outcomes including myocardial infarction, ischaemic stroke, cardiovascular death, and all-cause mortality.
This study developed CARDIAC-FM, a multimodal foundation model that integrates 12-lead ECG and cardiac MRI through self-supervised representation learning and cross-modal contrastive alignment, pretrained on 57,609 paired samples from UK Biobank, improving prediction of incident atrial fibrillation and heart failure, generalizing zero-shot to the Cardiovascular Health Study and the Multi-Ethnic Study of Atherosclerosis, and transferring to cardiac MRI phenotype prediction, time-to-event modelling, and additional outcomes including myocardial infarction, ischaemic stroke, cardiovascular death, and all-cause mortality.
JMIR Formative Research This exploratory study used GPT-4 to generate and iteratively fine-tune an R script for preprocessing and blob identification/counting in adaptive optics flood illumination ophthalmoscopy (AO-FIO) images, debugging it on images from 4 participants (1 healthy individual and 3 patients with Stargardt disease), having another researcher naive to the prior coding check the code for errors using a different test set of images from 4 other participants (1 healthy individual and 3 patients with Stargardt disease), and comparing cone counts from 5 AO image snippets with counts independently recorded by 2 human graders and those from pre-existing AO analysis software, with the authors positioning the script as functional but nonvalidated.
This exploratory study used GPT-4 to generate and iteratively fine-tune an R script for preprocessing and blob identification/counting in adaptive optics flood illumination ophthalmoscopy (AO-FIO) images, debugging it on images from 4 participants (1 healthy individual and 3 patients with Stargardt disease), having another researcher naive to the prior coding check the code for errors using a different test set of images from 4 other participants (1 healthy individual and 3 patients with Stargardt disease), and comparing cone counts from 5 AO image snippets with counts independently recorded by 2 human graders and those from pre-existing AO analysis software, with the authors positioning the script as functional but nonvalidated.
This exploratory study used GPT-4 to generate and iteratively fine-tune an R script for preprocessing and blob identification/counting in adaptive optics flood illumination ophthalmoscopy (AO-FIO) images, debugging it on images from 4 participants (1 healthy individual and 3 patients with Stargardt disease), having another researcher naive to the prior coding check the code for errors using a different test set of images from 4 other participants (1 healthy individual and 3 patients with Stargardt disease), and comparing cone counts from 5 AO image snippets with counts independently recorded by 2 human graders and those from pre-existing AO analysis software, with the authors positioning the script as functional but nonvalidated.
This exploratory study used GPT-4 to generate and iteratively fine-tune an R script for preprocessing and blob identification/counting in adaptive optics flood illumination ophthalmoscopy (AO-FIO) images, debugging it on images from 4 participants (1 healthy individual and 3 patients with Stargardt disease), having another researcher naive to the prior coding check the code for errors using a different test set of images from 4 other participants (1 healthy individual and 3 patients with Stargardt disease), and comparing cone counts from 5 AO image snippets with counts independently recorded by 2 human graders and those from pre-existing AO analysis software, with the authors positioning the script as functional but nonvalidated.
Skeletal Radiology This retrospective diagnostic accuracy study enrolled 539 consecutive patients aged 60 years or older undergoing hip and pelvis radiography at two emergency departments, used a consensus panel of three musculoskeletal radiologists and two senior orthopaedic trauma surgeons blinded to AI output as the reference standard, and evaluated Milvue SmartUrgences, which classified radiographs as "YES," "NO," or "DOUBT"; 487 patients received definitive classifications and entered the primary analysis (mean age 83.4 years, SD 8.4; 66% female; 191 hip fractures identified by the reference standard), yielding sensitivity 97.4% (95% CI 94.0-98.9), specificity 86.5% (95% CI 82.1-89.9), positive predictive value 82.3% (95% CI 76.8-86.7), negative predictive value 98.1% (95% CI 95.6-99.2), accuracy 90.
This retrospective diagnostic accuracy study enrolled 539 consecutive patients aged 60 years or older undergoing hip and pelvis radiography at two emergency departments, used a consensus panel of three musculoskeletal radiologists and two senior orthopaedic trauma surgeons blinded to AI output as the reference standard, and evaluated Milvue SmartUrgences, which classified radiographs as "YES," "NO," or "DOUBT"; 487 patients received definitive classifications and entered the primary analysis (mean age 83.4 years, SD 8.4; 66% female; 191 hip fractures identified by the reference standard), yielding sensitivity 97.4% (95% CI 94.0-98.9), specificity 86.5% (95% CI 82.1-89.9), positive predictive value 82.3% (95% CI 76.8-86.7), negative predictive value 98.1% (95% CI 95.6-99.2), accuracy 90.
This retrospective diagnostic accuracy study enrolled 539 consecutive patients aged 60 years or older undergoing hip and pelvis radiography at two emergency departments, used a consensus panel of three musculoskeletal radiologists and two senior orthopaedic trauma surgeons blinded to AI output as the reference standard, and evaluated Milvue SmartUrgences, which classified radiographs as "YES," "NO," or "DOUBT"; 487 patients received definitive classifications and entered the primary analysis (mean age 83.4 years, SD 8.4; 66% female; 191 hip fractures identified by the reference standard), yielding sensitivity 97.4% (95% CI 94.0-98.9), specificity 86.5% (95% CI 82.1-89.9), positive predictive value 82.3% (95% CI 76.8-86.7), negative predictive value 98.1% (95% CI 95.6-99.2), accuracy 90.
This retrospective diagnostic accuracy study enrolled 539 consecutive patients aged 60 years or older undergoing hip and pelvis radiography at two emergency departments, used a consensus panel of three musculoskeletal radiologists and two senior orthopaedic trauma surgeons blinded to AI output as the reference standard, and evaluated Milvue SmartUrgences, which classified radiographs as "YES," "NO," or "DOUBT"; 487 patients received definitive classifications and entered the primary analysis (mean age 83.4 years, SD 8.4; 66% female; 191 hip fractures identified by the reference standard), yielding sensitivity 97.4% (95% CI 94.0-98.9), specificity 86.5% (95% CI 82.1-89.9), positive predictive value 82.3% (95% CI 76.8-86.7), negative predictive value 98.1% (95% CI 95.6-99.2), accuracy 90.
Journal of Infection and Chemotherapy This pilot study collected ticks between May 2023 and December 2024 from patients presenting tick bites at 10 medical institutions in Okayama, Hiroshima, and Kagawa prefectures, supplemented with wild tick images, and developed a two-stage AI pipeline of object detection and genus-level classification in which a YOLO-based model trained on 3258 annotated public images achieved mAP@0.5 of 0.954 and ResNet50 achieved mean validation accuracy of 95% on 533 tick images covering four genera, integrating the system into a prototype smartphone application with geospatial visualization to support clinical risk assessment and public health surveillance.
This pilot study collected ticks between May 2023 and December 2024 from patients presenting tick bites at 10 medical institutions in Okayama, Hiroshima, and Kagawa prefectures, supplemented with wild tick images, and developed a two-stage AI pipeline of object detection and genus-level classification in which a YOLO-based model trained on 3258 annotated public images achieved mAP@0.5 of 0.954 and ResNet50 achieved mean validation accuracy of 95% on 533 tick images covering four genera, integrating the system into a prototype smartphone application with geospatial visualization to support clinical risk assessment and public health surveillance.
This pilot study collected ticks between May 2023 and December 2024 from patients presenting tick bites at 10 medical institutions in Okayama, Hiroshima, and Kagawa prefectures, supplemented with wild tick images, and developed a two-stage AI pipeline of object detection and genus-level classification in which a YOLO-based model trained on 3258 annotated public images achieved mAP@0.5 of 0.954 and ResNet50 achieved mean validation accuracy of 95% on 533 tick images covering four genera, integrating the system into a prototype smartphone application with geospatial visualization to support clinical risk assessment and public health surveillance.
This pilot study collected ticks between May 2023 and December 2024 from patients presenting tick bites at 10 medical institutions in Okayama, Hiroshima, and Kagawa prefectures, supplemented with wild tick images, and developed a two-stage AI pipeline of object detection and genus-level classification in which a YOLO-based model trained on 3258 annotated public images achieved mAP@0.5 of 0.954 and ResNet50 achieved mean validation accuracy of 95% on 533 tick images covering four genera, integrating the system into a prototype smartphone application with geospatial visualization to support clinical risk assessment and public health surveillance.
npj Digital Medicine This work introduces BenchECG, a standardized benchmark spanning eight public ECG datasets, 421,171 patients, 1,674,704 recordings, and ten tasks (classification, segmentation, detection, regression, survival analysis), used to evaluate public foundation models such as ST-MEM, ECG-JEPA, and ECGFounder; it also proposes xECG, a bidirectional xLSTM model pretrained with SimDINOv2 self-supervised learning, which achieves the best BenchECG score of 0.868±0.0030 (mean rank 1.50 under finetuning and 1.20 under linear probing), is the only public model to perform strongly across all datasets and task types, and leads on long-context tasks (sleep apnea AUROC 0.932±0.014; MIT-BIH arrhythmia F1 0.677±0.025) and computational efficiency (about 10x less time and about 7x less memory on PTB-XL).
This work introduces BenchECG, a standardized benchmark spanning eight public ECG datasets, 421,171 patients, 1,674,704 recordings, and ten tasks (classification, segmentation, detection, regression, survival analysis), used to evaluate public foundation models such as ST-MEM, ECG-JEPA, and ECGFounder; it also proposes xECG, a bidirectional xLSTM model pretrained with SimDINOv2 self-supervised learning, which achieves the best BenchECG score of 0.868±0.0030 (mean rank 1.50 under finetuning and 1.20 under linear probing), is the only public model to perform strongly across all datasets and task types, and leads on long-context tasks (sleep apnea AUROC 0.932±0.014; MIT-BIH arrhythmia F1 0.677±0.025) and computational efficiency (about 10x less time and about 7x less memory on PTB-XL).
This work introduces BenchECG, a standardized benchmark spanning eight public ECG datasets, 421,171 patients, 1,674,704 recordings, and ten tasks (classification, segmentation, detection, regression, survival analysis), used to evaluate public foundation models such as ST-MEM, ECG-JEPA, and ECGFounder; it also proposes xECG, a bidirectional xLSTM model pretrained with SimDINOv2 self-supervised learning, which achieves the best BenchECG score of 0.868±0.0030 (mean rank 1.50 under finetuning and 1.20 under linear probing), is the only public model to perform strongly across all datasets and task types, and leads on long-context tasks (sleep apnea AUROC 0.932±0.014; MIT-BIH arrhythmia F1 0.677±0.025) and computational efficiency (about 10x less time and about 7x less memory on PTB-XL).
This work introduces BenchECG, a standardized benchmark spanning eight public ECG datasets, 421,171 patients, 1,674,704 recordings, and ten tasks (classification, segmentation, detection, regression, survival analysis), used to evaluate public foundation models such as ST-MEM, ECG-JEPA, and ECGFounder; it also proposes xECG, a bidirectional xLSTM model pretrained with SimDINOv2 self-supervised learning, which achieves the best BenchECG score of 0.868±0.0030 (mean rank 1.50 under finetuning and 1.20 under linear probing), is the only public model to perform strongly across all datasets and task types, and leads on long-context tasks (sleep apnea AUROC 0.932±0.014; MIT-BIH arrhythmia F1 0.677±0.025) and computational efficiency (about 10x less time and about 7x less memory on PTB-XL).
Journal of Chemical Information and Modeling This study comprehensively benchmarked tabular foundation models (TFMs) based on in-context learning for small organic molecule property prediction, comparing several TFMs with multiple machine learning methods across 11 data sets (regression, random and structure-aware splits, up to 10,000 molecules each), and found that TFMs consistently outperform XGBoost, CatBoost, multilayer perceptrons, and other descriptor-based methods even with careful hyperparameter selection for the baselines, achieve accuracy on par with or better than graph-based methods including those pretrained on chemical data, with Uni-Mol2 slightly outperforming TFMs in some experiments, while retrieval (selecting the 500 closest neighbors by Tanimoto similarity for each test molecule) yields further improvement on relat
This study comprehensively benchmarked tabular foundation models (TFMs) based on in-context learning for small organic molecule property prediction, comparing several TFMs with multiple machine learning methods across 11 data sets (regression, random and structure-aware splits, up to 10,000 molecules each), and found that TFMs consistently outperform XGBoost, CatBoost, multilayer perceptrons, and other descriptor-based methods even with careful hyperparameter selection for the baselines, achieve accuracy on par with or better than graph-based methods including those pretrained on chemical data, with Uni-Mol2 slightly outperforming TFMs in some experiments, while retrieval (selecting the 500 closest neighbors by Tanimoto similarity for each test molecule) yields further improvement on relat
This study comprehensively benchmarked tabular foundation models (TFMs) based on in-context learning for small organic molecule property prediction, comparing several TFMs with multiple machine learning methods across 11 data sets (regression, random and structure-aware splits, up to 10,000 molecules each), and found that TFMs consistently outperform XGBoost, CatBoost, multilayer perceptrons, and other descriptor-based methods even with careful hyperparameter selection for the baselines, achieve accuracy on par with or better than graph-based methods including those pretrained on chemical data, with Uni-Mol2 slightly outperforming TFMs in some experiments, while retrieval (selecting the 500 closest neighbors by Tanimoto similarity for each test molecule) yields further improvement on relat
This study comprehensively benchmarked tabular foundation models (TFMs) based on in-context learning for small organic molecule property prediction, comparing several TFMs with multiple machine learning methods across 11 data sets (regression, random and structure-aware splits, up to 10,000 molecules each), and found that TFMs consistently outperform XGBoost, CatBoost, multilayer perceptrons, and other descriptor-based methods even with careful hyperparameter selection for the baselines, achieve accuracy on par with or better than graph-based methods including those pretrained on chemical data, with Uni-Mol2 slightly outperforming TFMs in some experiments, while retrieval (selecting the 500 closest neighbors by Tanimoto similarity for each test molecule) yields further improvement on relat
Cineca Institutional Research Information System (Tor Vergata University) This study devised a bioinformatics pipeline for a comparative analysis of mechanosensitive membrane proteins in lactic acid bacteria (Streptococcus thermophilus and Lactobacillus delbrueckii subsp. bulgaricus) and a fungus (Pleurotus floridanus), generating a consensus sequence for P. floridanus via multiple sequence alignment of homologues and mapping it onto the genome with BLAST to delineate the gene region of interest, then producing protein structures with the AlphaFold3 artificial intelligence platform using the correct oligomeric state for each channel, thereby bridging the structural knowledge gap for mechanoreceptors in these non-model systems.
This study devised a bioinformatics pipeline for a comparative analysis of mechanosensitive membrane proteins in lactic acid bacteria (Streptococcus thermophilus and Lactobacillus delbrueckii subsp. bulgaricus) and a fungus (Pleurotus floridanus), generating a consensus sequence for P. floridanus via multiple sequence alignment of homologues and mapping it onto the genome with BLAST to delineate the gene region of interest, then producing protein structures with the AlphaFold3 artificial intelligence platform using the correct oligomeric state for each channel, thereby bridging the structural knowledge gap for mechanoreceptors in these non-model systems.
This study devised a bioinformatics pipeline for a comparative analysis of mechanosensitive membrane proteins in lactic acid bacteria (Streptococcus thermophilus and Lactobacillus delbrueckii subsp. bulgaricus) and a fungus (Pleurotus floridanus), generating a consensus sequence for P. floridanus via multiple sequence alignment of homologues and mapping it onto the genome with BLAST to delineate the gene region of interest, then producing protein structures with the AlphaFold3 artificial intelligence platform using the correct oligomeric state for each channel, thereby bridging the structural knowledge gap for mechanoreceptors in these non-model systems.
This study devised a bioinformatics pipeline for a comparative analysis of mechanosensitive membrane proteins in lactic acid bacteria (Streptococcus thermophilus and Lactobacillus delbrueckii subsp. bulgaricus) and a fungus (Pleurotus floridanus), generating a consensus sequence for P. floridanus via multiple sequence alignment of homologues and mapping it onto the genome with BLAST to delineate the gene region of interest, then producing protein structures with the AlphaFold3 artificial intelligence platform using the correct oligomeric state for each channel, thereby bridging the structural knowledge gap for mechanoreceptors in these non-model systems.
Physics in Medicine and Biology In a training cohort (n = 186), a validation cohort (n = 116), and a prospective multicenter validation cohort (n = 269), this study extracted radiomic and supervised deep learning features from pretreatment T2-weighted MRI and compared manually developed with GPT-assisted modeling workflows, finding that fused features achieved the highest AUC for both manually developed and GPT-assisted logistic regression models (0.759 versus 0.763 ± 0.003 in the training cohort, 0.714 versus 0.741 ± 0.008 in the validation cohort, and 0.700 versus 0.706 ± 0.
In a training cohort (n = 186), a validation cohort (n = 116), and a prospective multicenter validation cohort (n = 269), this study extracted radiomic and supervised deep learning features from pretreatment T2-weighted MRI and compared manually developed with GPT-assisted modeling workflows, finding that fused features achieved the highest AUC for both manually developed and GPT-assisted logistic regression models (0.759 versus 0.763 ± 0.003 in the training cohort, 0.714 versus 0.741 ± 0.008 in the validation cohort, and 0.700 versus 0.706 ± 0.
In a training cohort (n = 186), a validation cohort (n = 116), and a prospective multicenter validation cohort (n = 269), this study extracted radiomic and supervised deep learning features from pretreatment T2-weighted MRI and compared manually developed with GPT-assisted modeling workflows, finding that fused features achieved the highest AUC for both manually developed and GPT-assisted logistic regression models (0.759 versus 0.763 ± 0.003 in the training cohort, 0.714 versus 0.741 ± 0.008 in the validation cohort, and 0.700 versus 0.706 ± 0.
In a training cohort (n = 186), a validation cohort (n = 116), and a prospective multicenter validation cohort (n = 269), this study extracted radiomic and supervised deep learning features from pretreatment T2-weighted MRI and compared manually developed with GPT-assisted modeling workflows, finding that fused features achieved the highest AUC for both manually developed and GPT-assisted logistic regression models (0.759 versus 0.763 ± 0.003 in the training cohort, 0.714 versus 0.741 ± 0.008 in the validation cohort, and 0.700 versus 0.706 ± 0.
World Journal of Otorhinolaryngology - Head and Neck Surgery Following PRISMA-ScR guidance, this scoping review searched EMBASE, PubMed (MEDLINE), CINAHL, SCOPUS, and Cochrane plus citation searching, and from 252 records included 12 studies, of which only 3 provided original data (one urology primary study, one urology expert commentary, and one narrative review with hand surgery survey data), with the remainder mostly narrative reviews whose cited evidence largely came from nonsurgical specialties, indicating that empirical research on ambient dictation AI in surgical specialties remains sparse.
Following PRISMA-ScR guidance, this scoping review searched EMBASE, PubMed (MEDLINE), CINAHL, SCOPUS, and Cochrane plus citation searching, and from 252 records included 12 studies, of which only 3 provided original data (one urology primary study, one urology expert commentary, and one narrative review with hand surgery survey data), with the remainder mostly narrative reviews whose cited evidence largely came from nonsurgical specialties, indicating that empirical research on ambient dictation AI in surgical specialties remains sparse.
Following PRISMA-ScR guidance, this scoping review searched EMBASE, PubMed (MEDLINE), CINAHL, SCOPUS, and Cochrane plus citation searching, and from 252 records included 12 studies, of which only 3 provided original data (one urology primary study, one urology expert commentary, and one narrative review with hand surgery survey data), with the remainder mostly narrative reviews whose cited evidence largely came from nonsurgical specialties, indicating that empirical research on ambient dictation AI in surgical specialties remains sparse.
Following PRISMA-ScR guidance, this scoping review searched EMBASE, PubMed (MEDLINE), CINAHL, SCOPUS, and Cochrane plus citation searching, and from 252 records included 12 studies, of which only 3 provided original data (one urology primary study, one urology expert commentary, and one narrative review with hand surgery survey data), with the remainder mostly narrative reviews whose cited evidence largely came from nonsurgical specialties, indicating that empirical research on ambient dictation AI in surgical specialties remains sparse.
arXiv This work introduces ResonatorLM, which replaces self-attention with a causal resonant field mixer built from damped resonators, treating a token sequence as a driven latent field, running training and prefill through O(T log T) causal FFT convolution and decoding through a fixed-size recurrent state; in a matched setting near 6M parameters, WikiText-2 character-level test perplexity drops from 4.617 to 3.764 and accuracy rises from 55.32% to 61.31%, decode reaches a 6.47x speedup over an optimized attention block at 32K tokens, and a separate kernel-level benchmark reports 440.29x at 8K and 575.86x at 32K.
This work introduces ResonatorLM, which replaces self-attention with a causal resonant field mixer built from damped resonators, treating a token sequence as a driven latent field, running training and prefill through O(T log T) causal FFT convolution and decoding through a fixed-size recurrent state; in a matched setting near 6M parameters, WikiText-2 character-level test perplexity drops from 4.617 to 3.764 and accuracy rises from 55.32% to 61.31%, decode reaches a 6.47x speedup over an optimized attention block at 32K tokens, and a separate kernel-level benchmark reports 440.29x at 8K and 575.86x at 32K.
This work introduces ResonatorLM, which replaces self-attention with a causal resonant field mixer built from damped resonators, treating a token sequence as a driven latent field, running training and prefill through O(T log T) causal FFT convolution and decoding through a fixed-size recurrent state; in a matched setting near 6M parameters, WikiText-2 character-level test perplexity drops from 4.617 to 3.764 and accuracy rises from 55.32% to 61.31%, decode reaches a 6.47x speedup over an optimized attention block at 32K tokens, and a separate kernel-level benchmark reports 440.29x at 8K and 575.86x at 32K.
This work introduces ResonatorLM, which replaces self-attention with a causal resonant field mixer built from damped resonators, treating a token sequence as a driven latent field, running training and prefill through O(T log T) causal FFT convolution and decoding through a fixed-size recurrent state; in a matched setting near 6M parameters, WikiText-2 character-level test perplexity drops from 4.617 to 3.764 and accuracy rises from 55.32% to 61.31%, decode reaches a 6.47x speedup over an optimized attention block at 32K tokens, and a separate kernel-level benchmark reports 440.29x at 8K and 575.86x at 32K.
Journal of Chemical Information and Modeling This paper presents a method for autogenerating domain-specific question-answering data, builds MagQA with 168,080 magnetic-materials QA pairs, and uses it to finetune BERT-style models; on a manually annotated magnetic-materials test set, a vanilla BERT-base-cased finetuned on MagQA mixed with SQuAD v2 (MagBERT_MagQA_Mixed) achieves the best result with an F1 of 78.43% and an exact-match score of 72.84%, indicating that, given sufficiently large and high-quality domain-specific QA data, domain-specific BERT models need only be finetuned from vanilla BERT without domain-adaptive pretraining.
This paper presents a method for autogenerating domain-specific question-answering data, builds MagQA with 168,080 magnetic-materials QA pairs, and uses it to finetune BERT-style models; on a manually annotated magnetic-materials test set, a vanilla BERT-base-cased finetuned on MagQA mixed with SQuAD v2 (MagBERT_MagQA_Mixed) achieves the best result with an F1 of 78.43% and an exact-match score of 72.84%, indicating that, given sufficiently large and high-quality domain-specific QA data, domain-specific BERT models need only be finetuned from vanilla BERT without domain-adaptive pretraining.
This paper presents a method for autogenerating domain-specific question-answering data, builds MagQA with 168,080 magnetic-materials QA pairs, and uses it to finetune BERT-style models; on a manually annotated magnetic-materials test set, a vanilla BERT-base-cased finetuned on MagQA mixed with SQuAD v2 (MagBERT_MagQA_Mixed) achieves the best result with an F1 of 78.43% and an exact-match score of 72.84%, indicating that, given sufficiently large and high-quality domain-specific QA data, domain-specific BERT models need only be finetuned from vanilla BERT without domain-adaptive pretraining.
This paper presents a method for autogenerating domain-specific question-answering data, builds MagQA with 168,080 magnetic-materials QA pairs, and uses it to finetune BERT-style models; on a manually annotated magnetic-materials test set, a vanilla BERT-base-cased finetuned on MagQA mixed with SQuAD v2 (MagBERT_MagQA_Mixed) achieves the best result with an F1 of 78.43% and an exact-match score of 72.84%, indicating that, given sufficiently large and high-quality domain-specific QA data, domain-specific BERT models need only be finetuned from vanilla BERT without domain-adaptive pretraining.
Journal of Chemical Information and Modeling The work presents LinkLlama, a fine-tuned Meta Llama 3 model supervised on a curated corpus of drug-like molecules from ChEMBL that accepts natural language prompts specifying geometric constraints such as distances and angles alongside physicochemical targets like Lipinski's rules and rotatable bond limits to generate tailored molecules for input fragments; benchmarking on the ZINC and HiQBind data sets shows competitive geometric fidelity relative to strictly 3D-aware models together with an approximately two-fold increase in the proportion of chemically reasonable designs, rising from 35% to over 80% under structural filters including PAINS, non-drug-like chemical patterns, and complex ring systems, with versatility illustrated through small-molecule scaffold hopping and PROTAC linker d
The work presents LinkLlama, a fine-tuned Meta Llama 3 model supervised on a curated corpus of drug-like molecules from ChEMBL that accepts natural language prompts specifying geometric constraints such as distances and angles alongside physicochemical targets like Lipinski's rules and rotatable bond limits to generate tailored molecules for input fragments; benchmarking on the ZINC and HiQBind data sets shows competitive geometric fidelity relative to strictly 3D-aware models together with an approximately two-fold increase in the proportion of chemically reasonable designs, rising from 35% to over 80% under structural filters including PAINS, non-drug-like chemical patterns, and complex ring systems, with versatility illustrated through small-molecule scaffold hopping and PROTAC linker d
The work presents LinkLlama, a fine-tuned Meta Llama 3 model supervised on a curated corpus of drug-like molecules from ChEMBL that accepts natural language prompts specifying geometric constraints such as distances and angles alongside physicochemical targets like Lipinski's rules and rotatable bond limits to generate tailored molecules for input fragments; benchmarking on the ZINC and HiQBind data sets shows competitive geometric fidelity relative to strictly 3D-aware models together with an approximately two-fold increase in the proportion of chemically reasonable designs, rising from 35% to over 80% under structural filters including PAINS, non-drug-like chemical patterns, and complex ring systems, with versatility illustrated through small-molecule scaffold hopping and PROTAC linker d
The work presents LinkLlama, a fine-tuned Meta Llama 3 model supervised on a curated corpus of drug-like molecules from ChEMBL that accepts natural language prompts specifying geometric constraints such as distances and angles alongside physicochemical targets like Lipinski's rules and rotatable bond limits to generate tailored molecules for input fragments; benchmarking on the ZINC and HiQBind data sets shows competitive geometric fidelity relative to strictly 3D-aware models together with an approximately two-fold increase in the proportion of chemically reasonable designs, rising from 35% to over 80% under structural filters including PAINS, non-drug-like chemical patterns, and complex ring systems, with versatility illustrated through small-molecule scaffold hopping and PROTAC linker d
JMIR Infodemiology Analyzing 27,434 posts from the Reddit SuicideWatch forum with a combination of manual review, large language model-assisted keyword expansion, natural language processing-based information extraction, and human validation, this study identified posts where suicidal ideation and family abuse co-occurred and described self-disclosed age and gender, abuse types, and relationships to perpetrators: among posts with self-reported age, individuals aged 18 to 24 accounted for the largest proportion (144/313, 46.0%); among posts with self-reported gender, men represented 57.3% (86/150); physical abuse was most frequent (683/975, 70.1%; 95% CI 67.18%-72.93%), followed by emotional or psychological abuse (350/975, 35.9%; 95% CI 32.89%-38.91%) and sexual abuse (296/975, 30.4%; 95% CI 27.47%-33.
Analyzing 27,434 posts from the Reddit SuicideWatch forum with a combination of manual review, large language model-assisted keyword expansion, natural language processing-based information extraction, and human validation, this study identified posts where suicidal ideation and family abuse co-occurred and described self-disclosed age and gender, abuse types, and relationships to perpetrators: among posts with self-reported age, individuals aged 18 to 24 accounted for the largest proportion (144/313, 46.0%); among posts with self-reported gender, men represented 57.3% (86/150); physical abuse was most frequent (683/975, 70.1%; 95% CI 67.18%-72.93%), followed by emotional or psychological abuse (350/975, 35.9%; 95% CI 32.89%-38.91%) and sexual abuse (296/975, 30.4%; 95% CI 27.47%-33.
Analyzing 27,434 posts from the Reddit SuicideWatch forum with a combination of manual review, large language model-assisted keyword expansion, natural language processing-based information extraction, and human validation, this study identified posts where suicidal ideation and family abuse co-occurred and described self-disclosed age and gender, abuse types, and relationships to perpetrators: among posts with self-reported age, individuals aged 18 to 24 accounted for the largest proportion (144/313, 46.0%); among posts with self-reported gender, men represented 57.3% (86/150); physical abuse was most frequent (683/975, 70.1%; 95% CI 67.18%-72.93%), followed by emotional or psychological abuse (350/975, 35.9%; 95% CI 32.89%-38.91%) and sexual abuse (296/975, 30.4%; 95% CI 27.47%-33.
Analyzing 27,434 posts from the Reddit SuicideWatch forum with a combination of manual review, large language model-assisted keyword expansion, natural language processing-based information extraction, and human validation, this study identified posts where suicidal ideation and family abuse co-occurred and described self-disclosed age and gender, abuse types, and relationships to perpetrators: among posts with self-reported age, individuals aged 18 to 24 accounted for the largest proportion (144/313, 46.0%); among posts with self-reported gender, men represented 57.3% (86/150); physical abuse was most frequent (683/975, 70.1%; 95% CI 67.18%-72.93%), followed by emotional or psychological abuse (350/975, 35.9%; 95% CI 32.89%-38.91%) and sexual abuse (296/975, 30.4%; 95% CI 27.47%-33.
Journal of Chemical Information and Modeling Drawing on two case studies characterized by structural and biochemical data, this article examines how the close integration of computational and medicinal chemistry, together with a deep understanding of chemical shape and protein interactions, enabled targeted computational molecular design to produce disproportionate gains in efficacy and selectivity, and on that basis argues that current AI methods are sensitive to subtle, low-data perturbations, that AI must move beyond pattern recognition to understand or explicitly simulate the mechanistic "why" linking subtle structural changes to biological outcomes, and that until then AI is better positioned to complement human efforts than to serve as a stand-alone solution.
Drawing on two case studies characterized by structural and biochemical data, this article examines how the close integration of computational and medicinal chemistry, together with a deep understanding of chemical shape and protein interactions, enabled targeted computational molecular design to produce disproportionate gains in efficacy and selectivity, and on that basis argues that current AI methods are sensitive to subtle, low-data perturbations, that AI must move beyond pattern recognition to understand or explicitly simulate the mechanistic "why" linking subtle structural changes to biological outcomes, and that until then AI is better positioned to complement human efforts than to serve as a stand-alone solution.
Drawing on two case studies characterized by structural and biochemical data, this article examines how the close integration of computational and medicinal chemistry, together with a deep understanding of chemical shape and protein interactions, enabled targeted computational molecular design to produce disproportionate gains in efficacy and selectivity, and on that basis argues that current AI methods are sensitive to subtle, low-data perturbations, that AI must move beyond pattern recognition to understand or explicitly simulate the mechanistic "why" linking subtle structural changes to biological outcomes, and that until then AI is better positioned to complement human efforts than to serve as a stand-alone solution.
Drawing on two case studies characterized by structural and biochemical data, this article examines how the close integration of computational and medicinal chemistry, together with a deep understanding of chemical shape and protein interactions, enabled targeted computational molecular design to produce disproportionate gains in efficacy and selectivity, and on that basis argues that current AI methods are sensitive to subtle, low-data perturbations, that AI must move beyond pattern recognition to understand or explicitly simulate the mechanistic "why" linking subtle structural changes to biological outcomes, and that until then AI is better positioned to complement human efforts than to serve as a stand-alone solution.
Academic Radiology This article argues that radiology should treat sequence as a design variable in routine clinical practice: capture what the radiologist concluded before AI exposure, then what the AI displayed and how the radiologist responded, thereby preserving the pre-AI interpretation as an auditable event, while noting that responsibility remains with the radiologist who signs the report and that the proposal is bounded by what each device is cleared to do.
This article argues that radiology should treat sequence as a design variable in routine clinical practice: capture what the radiologist concluded before AI exposure, then what the AI displayed and how the radiologist responded, thereby preserving the pre-AI interpretation as an auditable event, while noting that responsibility remains with the radiologist who signs the report and that the proposal is bounded by what each device is cleared to do.
This article argues that radiology should treat sequence as a design variable in routine clinical practice: capture what the radiologist concluded before AI exposure, then what the AI displayed and how the radiologist responded, thereby preserving the pre-AI interpretation as an auditable event, while noting that responsibility remains with the radiologist who signs the report and that the proposal is bounded by what each device is cleared to do.
This article argues that radiology should treat sequence as a design variable in routine clinical practice: capture what the radiologist concluded before AI exposure, then what the AI displayed and how the radiologist responded, thereby preserving the pre-AI interpretation as an auditable event, while noting that responsibility remains with the radiologist who signs the report and that the proposal is bounded by what each device is cleared to do.
Journal of Medical Internet Research This study built a gold-standard corpus of 1000 Portuguese outpatient clinical notes manually annotated by 5 trained researchers for 5 protected-entity categories (patient names, dates, identifiers, organizations, and geographic locations) and, on a held-out test set of 500 notes, compared two stand-alone LLMs (Llama-3.1-8B-instruct and Llama-3.3-70B-instruct) with two quantum-enhanced hybrid models (Dynex-QML with 8B and 70B base models, using QUBO formulations to transform the final attention layer into a global constraint satisfaction problem solved by neuromorphic quantum annealing); the quantum-enhanced Dynex-QML-70B achieved the highest macro-F1 of 0.855 (95% CI 0.823-0.880), above stand-alone Llama-3.3-70B (0.726), Dynex-QML-8B (0.733), and Llama-3.1-8B (0.
This study built a gold-standard corpus of 1000 Portuguese outpatient clinical notes manually annotated by 5 trained researchers for 5 protected-entity categories (patient names, dates, identifiers, organizations, and geographic locations) and, on a held-out test set of 500 notes, compared two stand-alone LLMs (Llama-3.1-8B-instruct and Llama-3.3-70B-instruct) with two quantum-enhanced hybrid models (Dynex-QML with 8B and 70B base models, using QUBO formulations to transform the final attention layer into a global constraint satisfaction problem solved by neuromorphic quantum annealing); the quantum-enhanced Dynex-QML-70B achieved the highest macro-F1 of 0.855 (95% CI 0.823-0.880), above stand-alone Llama-3.3-70B (0.726), Dynex-QML-8B (0.733), and Llama-3.1-8B (0.
This study built a gold-standard corpus of 1000 Portuguese outpatient clinical notes manually annotated by 5 trained researchers for 5 protected-entity categories (patient names, dates, identifiers, organizations, and geographic locations) and, on a held-out test set of 500 notes, compared two stand-alone LLMs (Llama-3.1-8B-instruct and Llama-3.3-70B-instruct) with two quantum-enhanced hybrid models (Dynex-QML with 8B and 70B base models, using QUBO formulations to transform the final attention layer into a global constraint satisfaction problem solved by neuromorphic quantum annealing); the quantum-enhanced Dynex-QML-70B achieved the highest macro-F1 of 0.855 (95% CI 0.823-0.880), above stand-alone Llama-3.3-70B (0.726), Dynex-QML-8B (0.733), and Llama-3.1-8B (0.
This study built a gold-standard corpus of 1000 Portuguese outpatient clinical notes manually annotated by 5 trained researchers for 5 protected-entity categories (patient names, dates, identifiers, organizations, and geographic locations) and, on a held-out test set of 500 notes, compared two stand-alone LLMs (Llama-3.1-8B-instruct and Llama-3.3-70B-instruct) with two quantum-enhanced hybrid models (Dynex-QML with 8B and 70B base models, using QUBO formulations to transform the final attention layer into a global constraint satisfaction problem solved by neuromorphic quantum annealing); the quantum-enhanced Dynex-QML-70B achieved the highest macro-F1 of 0.855 (95% CI 0.823-0.880), above stand-alone Llama-3.3-70B (0.726), Dynex-QML-8B (0.733), and Llama-3.1-8B (0.
Graefe s Archive for Clinical and Experimental Ophthalmology This retrospective cohort study of 80 treatment-naive neovascular age-related macular degeneration patients followed for at least 4 years estimated retinal age from fellow-eye fundus photographs using the publicly available Japan Ocular Imaging Registry deep learning model and computed raw retinal age gap (retinal age minus chronological age), finding no association between raw or age-corrected retinal age gap and total injection number in the count analysis, while in an exploratory threshold-based analysis the ≥ 24-injection group had higher raw retinal age gap than the < 24-injection group (4.19 vs. 0.66 years, P = 0.039) and each 1-year increase in raw retinal age gap was associated with requiring ≥ 24 injections (OR 1.11; 95% CI, 1.01-1.22; P = 0.
This retrospective cohort study of 80 treatment-naive neovascular age-related macular degeneration patients followed for at least 4 years estimated retinal age from fellow-eye fundus photographs using the publicly available Japan Ocular Imaging Registry deep learning model and computed raw retinal age gap (retinal age minus chronological age), finding no association between raw or age-corrected retinal age gap and total injection number in the count analysis, while in an exploratory threshold-based analysis the ≥ 24-injection group had higher raw retinal age gap than the < 24-injection group (4.19 vs. 0.66 years, P = 0.039) and each 1-year increase in raw retinal age gap was associated with requiring ≥ 24 injections (OR 1.11; 95% CI, 1.01-1.22; P = 0.
This retrospective cohort study of 80 treatment-naive neovascular age-related macular degeneration patients followed for at least 4 years estimated retinal age from fellow-eye fundus photographs using the publicly available Japan Ocular Imaging Registry deep learning model and computed raw retinal age gap (retinal age minus chronological age), finding no association between raw or age-corrected retinal age gap and total injection number in the count analysis, while in an exploratory threshold-based analysis the ≥ 24-injection group had higher raw retinal age gap than the < 24-injection group (4.19 vs. 0.66 years, P = 0.039) and each 1-year increase in raw retinal age gap was associated with requiring ≥ 24 injections (OR 1.11; 95% CI, 1.01-1.22; P = 0.
This retrospective cohort study of 80 treatment-naive neovascular age-related macular degeneration patients followed for at least 4 years estimated retinal age from fellow-eye fundus photographs using the publicly available Japan Ocular Imaging Registry deep learning model and computed raw retinal age gap (retinal age minus chronological age), finding no association between raw or age-corrected retinal age gap and total injection number in the count analysis, while in an exploratory threshold-based analysis the ≥ 24-injection group had higher raw retinal age gap than the < 24-injection group (4.19 vs. 0.66 years, P = 0.039) and each 1-year increase in raw retinal age gap was associated with requiring ≥ 24 injections (OR 1.11; 95% CI, 1.01-1.22; P = 0.
Journal of Medical Internet Research Using the single-word prompt "Depression," this study generated 100 videos across two access points of Sora 2 (consumer app, n=50; developer API, n=50), had two trained coders independently code narrative structure, visual environments, objects, figure demographics, and figure states, and extracted computational features (visual aesthetics, audio, semantic content, temporal dynamics) for comparison, finding a pronounced recovery bias in app-generated videos (78%, 39/50, featured arcs progressing from depressive states toward resolution versus 14%, 7/50, of API outputs), with app videos brightening over time (mean slope 2.90, SD 2.43 per second vs -0.18, SD 1.24 per second for the API; Cohen d=1.59; q<.001) and containing three times more motion (Cohen d=2.07; q<.
Using the single-word prompt "Depression," this study generated 100 videos across two access points of Sora 2 (consumer app, n=50; developer API, n=50), had two trained coders independently code narrative structure, visual environments, objects, figure demographics, and figure states, and extracted computational features (visual aesthetics, audio, semantic content, temporal dynamics) for comparison, finding a pronounced recovery bias in app-generated videos (78%, 39/50, featured arcs progressing from depressive states toward resolution versus 14%, 7/50, of API outputs), with app videos brightening over time (mean slope 2.90, SD 2.43 per second vs -0.18, SD 1.24 per second for the API; Cohen d=1.59; q<.001) and containing three times more motion (Cohen d=2.07; q<.
Using the single-word prompt "Depression," this study generated 100 videos across two access points of Sora 2 (consumer app, n=50; developer API, n=50), had two trained coders independently code narrative structure, visual environments, objects, figure demographics, and figure states, and extracted computational features (visual aesthetics, audio, semantic content, temporal dynamics) for comparison, finding a pronounced recovery bias in app-generated videos (78%, 39/50, featured arcs progressing from depressive states toward resolution versus 14%, 7/50, of API outputs), with app videos brightening over time (mean slope 2.90, SD 2.43 per second vs -0.18, SD 1.24 per second for the API; Cohen d=1.59; q<.001) and containing three times more motion (Cohen d=2.07; q<.
Using the single-word prompt "Depression," this study generated 100 videos across two access points of Sora 2 (consumer app, n=50; developer API, n=50), had two trained coders independently code narrative structure, visual environments, objects, figure demographics, and figure states, and extracted computational features (visual aesthetics, audio, semantic content, temporal dynamics) for comparison, finding a pronounced recovery bias in app-generated videos (78%, 39/50, featured arcs progressing from depressive states toward resolution versus 14%, 7/50, of API outputs), with app videos brightening over time (mean slope 2.90, SD 2.43 per second vs -0.18, SD 1.24 per second for the API; Cohen d=1.59; q<.001) and containing three times more motion (Cohen d=2.07; q<.
Journal of Medical Internet Research Using 16,447 clinical notes from 6,382 patients with mental health diagnoses in the MIMIC-IV database, this study labeled sentiment from patient, physician, and general perspectives with two large language models (DeepSeek-7B and Mistral-7B) and three lexicon-based tools (ClinSent-lexicon, TextBlob, and VADER), finding substantial directional change in sentiment trajectories among patients with multiple admissions, greater fluctuation in Discharge Instructions than in Brief Hospital Course notes, more balanced patient-perspective sentiment versus predominantly neutral physician and general perspectives, better alignment of LLMs with patient-centered annotations than lexicon-based methods, and significantly more negative discharge-note sentiment trajectories among patients who died within 3
Using 16,447 clinical notes from 6,382 patients with mental health diagnoses in the MIMIC-IV database, this study labeled sentiment from patient, physician, and general perspectives with two large language models (DeepSeek-7B and Mistral-7B) and three lexicon-based tools (ClinSent-lexicon, TextBlob, and VADER), finding substantial directional change in sentiment trajectories among patients with multiple admissions, greater fluctuation in Discharge Instructions than in Brief Hospital Course notes, more balanced patient-perspective sentiment versus predominantly neutral physician and general perspectives, better alignment of LLMs with patient-centered annotations than lexicon-based methods, and significantly more negative discharge-note sentiment trajectories among patients who died within 3
Using 16,447 clinical notes from 6,382 patients with mental health diagnoses in the MIMIC-IV database, this study labeled sentiment from patient, physician, and general perspectives with two large language models (DeepSeek-7B and Mistral-7B) and three lexicon-based tools (ClinSent-lexicon, TextBlob, and VADER), finding substantial directional change in sentiment trajectories among patients with multiple admissions, greater fluctuation in Discharge Instructions than in Brief Hospital Course notes, more balanced patient-perspective sentiment versus predominantly neutral physician and general perspectives, better alignment of LLMs with patient-centered annotations than lexicon-based methods, and significantly more negative discharge-note sentiment trajectories among patients who died within 3
Using 16,447 clinical notes from 6,382 patients with mental health diagnoses in the MIMIC-IV database, this study labeled sentiment from patient, physician, and general perspectives with two large language models (DeepSeek-7B and Mistral-7B) and three lexicon-based tools (ClinSent-lexicon, TextBlob, and VADER), finding substantial directional change in sentiment trajectories among patients with multiple admissions, greater fluctuation in Discharge Instructions than in Brief Hospital Course notes, more balanced patient-perspective sentiment versus predominantly neutral physician and general perspectives, better alignment of LLMs with patient-centered annotations than lexicon-based methods, and significantly more negative discharge-note sentiment trajectories among patients who died within 3
medRxiv The study developed and evaluated the Multimodal Anonymizer, a modular, locally deployable multi-agent framework integrating multimodal large language models, task-specific neural networks, and rule-based transformations; on benchmarks spanning text, tables, PDFs, imaging, metadata, filenames, audio, handwriting, and 3D imaging, its best local configuration (orchestrator Qwen3-VL-235B-A22B-Thinking) achieved 98.80% per-patient deidentification sensitivity (95%-CI 97.20; 100) and 99.60% critical clinical preservation (95%-CI 98.80; 100), reached 100% sensitivity and critical preservation on 250 local Charité partograms, and outperformed established tools across most modalities.
The study developed and evaluated the Multimodal Anonymizer, a modular, locally deployable multi-agent framework integrating multimodal large language models, task-specific neural networks, and rule-based transformations; on benchmarks spanning text, tables, PDFs, imaging, metadata, filenames, audio, handwriting, and 3D imaging, its best local configuration (orchestrator Qwen3-VL-235B-A22B-Thinking) achieved 98.80% per-patient deidentification sensitivity (95%-CI 97.20; 100) and 99.60% critical clinical preservation (95%-CI 98.80; 100), reached 100% sensitivity and critical preservation on 250 local Charité partograms, and outperformed established tools across most modalities.
The study developed and evaluated the Multimodal Anonymizer, a modular, locally deployable multi-agent framework integrating multimodal large language models, task-specific neural networks, and rule-based transformations; on benchmarks spanning text, tables, PDFs, imaging, metadata, filenames, audio, handwriting, and 3D imaging, its best local configuration (orchestrator Qwen3-VL-235B-A22B-Thinking) achieved 98.80% per-patient deidentification sensitivity (95%-CI 97.20; 100) and 99.60% critical clinical preservation (95%-CI 98.80; 100), reached 100% sensitivity and critical preservation on 250 local Charité partograms, and outperformed established tools across most modalities.
The study developed and evaluated the Multimodal Anonymizer, a modular, locally deployable multi-agent framework integrating multimodal large language models, task-specific neural networks, and rule-based transformations; on benchmarks spanning text, tables, PDFs, imaging, metadata, filenames, audio, handwriting, and 3D imaging, its best local configuration (orchestrator Qwen3-VL-235B-A22B-Thinking) achieved 98.80% per-patient deidentification sensitivity (95%-CI 97.20; 100) and 99.60% critical clinical preservation (95%-CI 98.80; 100), reached 100% sensitivity and critical preservation on 250 local Charité partograms, and outperformed established tools across most modalities.
Pharmacogenomics This review article states that pharmacogenomics explains individual drug-response variation at the genetic level while artificial intelligence improves diagnostic accuracy, risk assessment, and treatment strategy through analysis of large and complex clinical, genetic, and imaging datasets, and that combining the two can support more individualized cardiovascular care and better clinical decision-making, though data security, ethical concerns, and insufficient clinical validation still limit widespread adoption.
This review article states that pharmacogenomics explains individual drug-response variation at the genetic level while artificial intelligence improves diagnostic accuracy, risk assessment, and treatment strategy through analysis of large and complex clinical, genetic, and imaging datasets, and that combining the two can support more individualized cardiovascular care and better clinical decision-making, though data security, ethical concerns, and insufficient clinical validation still limit widespread adoption.
This review article states that pharmacogenomics explains individual drug-response variation at the genetic level while artificial intelligence improves diagnostic accuracy, risk assessment, and treatment strategy through analysis of large and complex clinical, genetic, and imaging datasets, and that combining the two can support more individualized cardiovascular care and better clinical decision-making, though data security, ethical concerns, and insufficient clinical validation still limit widespread adoption.
This review article states that pharmacogenomics explains individual drug-response variation at the genetic level while artificial intelligence improves diagnostic accuracy, risk assessment, and treatment strategy through analysis of large and complex clinical, genetic, and imaging datasets, and that combining the two can support more individualized cardiovascular care and better clinical decision-making, though data security, ethical concerns, and insufficient clinical validation still limit widespread adoption.
Pharmaceutical Medicine This article describes the transformation of AstraZeneca's local Marketing Companies Regulatory Affairs (MCRA) teams in Europe from a predominantly compliance-driven support function into a strategic partner in drug development and patient access, organized around three pillars (Launch Excellence, External Engagement and Advocacy, and Digitalisation and Process Optimisation) and a "One-Team" local-global regulatory culture, and proposes a contribution-based measurement framework to monitor progress.
This article describes the transformation of AstraZeneca's local Marketing Companies Regulatory Affairs (MCRA) teams in Europe from a predominantly compliance-driven support function into a strategic partner in drug development and patient access, organized around three pillars (Launch Excellence, External Engagement and Advocacy, and Digitalisation and Process Optimisation) and a "One-Team" local-global regulatory culture, and proposes a contribution-based measurement framework to monitor progress.
This article describes the transformation of AstraZeneca's local Marketing Companies Regulatory Affairs (MCRA) teams in Europe from a predominantly compliance-driven support function into a strategic partner in drug development and patient access, organized around three pillars (Launch Excellence, External Engagement and Advocacy, and Digitalisation and Process Optimisation) and a "One-Team" local-global regulatory culture, and proposes a contribution-based measurement framework to monitor progress.
This article describes the transformation of AstraZeneca's local Marketing Companies Regulatory Affairs (MCRA) teams in Europe from a predominantly compliance-driven support function into a strategic partner in drug development and patient access, organized around three pillars (Launch Excellence, External Engagement and Advocacy, and Digitalisation and Process Optimisation) and a "One-Team" local-global regulatory culture, and proposes a contribution-based measurement framework to monitor progress.
npj Digital Medicine Responding to Qian et al.'s finding that a single LLM can generate acceptable knowledge-based questions but struggles with higher-order reasoning, this work proposes MAID, a multi-agent architecture that decomposes item development into specialized agents for drafting, critique, and iterative adversarial refinement, and in a blinded paired-comparison evaluation of items aligned with China's National Medical Licensing Examination standards, 14 experts from 7 medical disciplines compared 25 matched MCQ pairs yielding 350 item-level observations under a two-alternative forced-choice design, with multi-agent outputs receiving 57.7% of expert preferences (95% CI [52.5%, 62.8%]) versus 42.3% for the single-model baseline (95% CI [37.2%, 47.5%]; χ²(1) = 8.33, p = 0.
Responding to Qian et al.'s finding that a single LLM can generate acceptable knowledge-based questions but struggles with higher-order reasoning, this work proposes MAID, a multi-agent architecture that decomposes item development into specialized agents for drafting, critique, and iterative adversarial refinement, and in a blinded paired-comparison evaluation of items aligned with China's National Medical Licensing Examination standards, 14 experts from 7 medical disciplines compared 25 matched MCQ pairs yielding 350 item-level observations under a two-alternative forced-choice design, with multi-agent outputs receiving 57.7% of expert preferences (95% CI [52.5%, 62.8%]) versus 42.3% for the single-model baseline (95% CI [37.2%, 47.5%]; χ²(1) = 8.33, p = 0.
Responding to Qian et al.'s finding that a single LLM can generate acceptable knowledge-based questions but struggles with higher-order reasoning, this work proposes MAID, a multi-agent architecture that decomposes item development into specialized agents for drafting, critique, and iterative adversarial refinement, and in a blinded paired-comparison evaluation of items aligned with China's National Medical Licensing Examination standards, 14 experts from 7 medical disciplines compared 25 matched MCQ pairs yielding 350 item-level observations under a two-alternative forced-choice design, with multi-agent outputs receiving 57.7% of expert preferences (95% CI [52.5%, 62.8%]) versus 42.3% for the single-model baseline (95% CI [37.2%, 47.5%]; χ²(1) = 8.33, p = 0.
Responding to Qian et al.'s finding that a single LLM can generate acceptable knowledge-based questions but struggles with higher-order reasoning, this work proposes MAID, a multi-agent architecture that decomposes item development into specialized agents for drafting, critique, and iterative adversarial refinement, and in a blinded paired-comparison evaluation of items aligned with China's National Medical Licensing Examination standards, 14 experts from 7 medical disciplines compared 25 matched MCQ pairs yielding 350 item-level observations under a two-alternative forced-choice design, with multi-agent outputs receiving 57.7% of expert preferences (95% CI [52.5%, 62.8%]) versus 42.3% for the single-model baseline (95% CI [37.2%, 47.5%]; χ²(1) = 8.33, p = 0.
Human gene therapy This Perspective article presents the InsightRP2 framework, an integrated translational strategy combining clinical data, artificial intelligence-supported imaging analysis, experimental disease modeling, and adeno-associated virus design, with the aim of facilitating development of a targeted gene therapy for RP2-associated retinitis pigmentosa.
This Perspective article presents the InsightRP2 framework, an integrated translational strategy combining clinical data, artificial intelligence-supported imaging analysis, experimental disease modeling, and adeno-associated virus design, with the aim of facilitating development of a targeted gene therapy for RP2-associated retinitis pigmentosa.
This Perspective article presents the InsightRP2 framework, an integrated translational strategy combining clinical data, artificial intelligence-supported imaging analysis, experimental disease modeling, and adeno-associated virus design, with the aim of facilitating development of a targeted gene therapy for RP2-associated retinitis pigmentosa.
This Perspective article presents the InsightRP2 framework, an integrated translational strategy combining clinical data, artificial intelligence-supported imaging analysis, experimental disease modeling, and adeno-associated virus design, with the aim of facilitating development of a targeted gene therapy for RP2-associated retinitis pigmentosa.
medRxiv Across 193,101 Russian- and Kazakh-language stroke discharge summaries, with test cohorts of 332 section cases, 149 medication cases (2,475 reference records), and 191 laboratory cases, this study compared a multilingual encoder, locally fine-tuned Qwen3-4B models, and zero-shot GPT-5.5 on entity detection versus complete-record assembly, finding section F1 of 0.919/0.926/0.932, GPT-5.5 leading drug-name detection (0.966 versus 0.940) while Qwen led normalized medication recovery (0.381 versus 0.311; difference 0.070, 95% CI 0.026–0.114) and laboratory quintuple F1 (0.892 versus 0.822), and showing that moving from curated sections to a raw-document cascade reduced medication recovery from 0.377 to 0.204 (single window) and 0.246 (all blocks) and laboratory quintuple F1 from 0.898 to 0.
Across 193,101 Russian- and Kazakh-language stroke discharge summaries, with test cohorts of 332 section cases, 149 medication cases (2,475 reference records), and 191 laboratory cases, this study compared a multilingual encoder, locally fine-tuned Qwen3-4B models, and zero-shot GPT-5.5 on entity detection versus complete-record assembly, finding section F1 of 0.919/0.926/0.932, GPT-5.5 leading drug-name detection (0.966 versus 0.940) while Qwen led normalized medication recovery (0.381 versus 0.311; difference 0.070, 95% CI 0.026–0.114) and laboratory quintuple F1 (0.892 versus 0.822), and showing that moving from curated sections to a raw-document cascade reduced medication recovery from 0.377 to 0.204 (single window) and 0.246 (all blocks) and laboratory quintuple F1 from 0.898 to 0.
Across 193,101 Russian- and Kazakh-language stroke discharge summaries, with test cohorts of 332 section cases, 149 medication cases (2,475 reference records), and 191 laboratory cases, this study compared a multilingual encoder, locally fine-tuned Qwen3-4B models, and zero-shot GPT-5.5 on entity detection versus complete-record assembly, finding section F1 of 0.919/0.926/0.932, GPT-5.5 leading drug-name detection (0.966 versus 0.940) while Qwen led normalized medication recovery (0.381 versus 0.311; difference 0.070, 95% CI 0.026–0.114) and laboratory quintuple F1 (0.892 versus 0.822), and showing that moving from curated sections to a raw-document cascade reduced medication recovery from 0.377 to 0.204 (single window) and 0.246 (all blocks) and laboratory quintuple F1 from 0.898 to 0.
Across 193,101 Russian- and Kazakh-language stroke discharge summaries, with test cohorts of 332 section cases, 149 medication cases (2,475 reference records), and 191 laboratory cases, this study compared a multilingual encoder, locally fine-tuned Qwen3-4B models, and zero-shot GPT-5.5 on entity detection versus complete-record assembly, finding section F1 of 0.919/0.926/0.932, GPT-5.5 leading drug-name detection (0.966 versus 0.940) while Qwen led normalized medication recovery (0.381 versus 0.311; difference 0.070, 95% CI 0.026–0.114) and laboratory quintuple F1 (0.892 versus 0.822), and showing that moving from curated sections to a raw-document cascade reduced medication recovery from 0.377 to 0.204 (single window) and 0.246 (all blocks) and laboratory quintuple F1 from 0.898 to 0.
Journal of Chemical Information and Modeling This review systematically surveys AI-driven drug-target interaction (DTI) prediction, starting from classical molecular binding theories (lock-and-key, induced fit, conformational selection) and summarizing task settings such as binary interaction classification, binding affinity regression, and multitask prediction with uncertainty assessment; it organizes multimodal representations for drugs and target proteins (molecular sequences, graph structures, 3D conformations, physicochemical properties, biological perturbation profiles, protein sequences and structures, biomedical knowledge networks) together with interaction labels and auxiliary biomedical data, compares representative approaches across orthogonal dimensions including input representation, encoder architecture, interaction-mod
This review systematically surveys AI-driven drug-target interaction (DTI) prediction, starting from classical molecular binding theories (lock-and-key, induced fit, conformational selection) and summarizing task settings such as binary interaction classification, binding affinity regression, and multitask prediction with uncertainty assessment; it organizes multimodal representations for drugs and target proteins (molecular sequences, graph structures, 3D conformations, physicochemical properties, biological perturbation profiles, protein sequences and structures, biomedical knowledge networks) together with interaction labels and auxiliary biomedical data, compares representative approaches across orthogonal dimensions including input representation, encoder architecture, interaction-mod
This review systematically surveys AI-driven drug-target interaction (DTI) prediction, starting from classical molecular binding theories (lock-and-key, induced fit, conformational selection) and summarizing task settings such as binary interaction classification, binding affinity regression, and multitask prediction with uncertainty assessment; it organizes multimodal representations for drugs and target proteins (molecular sequences, graph structures, 3D conformations, physicochemical properties, biological perturbation profiles, protein sequences and structures, biomedical knowledge networks) together with interaction labels and auxiliary biomedical data, compares representative approaches across orthogonal dimensions including input representation, encoder architecture, interaction-mod
This review systematically surveys AI-driven drug-target interaction (DTI) prediction, starting from classical molecular binding theories (lock-and-key, induced fit, conformational selection) and summarizing task settings such as binary interaction classification, binding affinity regression, and multitask prediction with uncertainty assessment; it organizes multimodal representations for drugs and target proteins (molecular sequences, graph structures, 3D conformations, physicochemical properties, biological perturbation profiles, protein sequences and structures, biomedical knowledge networks) together with interaction labels and auxiliary biomedical data, compares representative approaches across orthogonal dimensions including input representation, encoder architecture, interaction-mod
Acta Pharmacologica Sinica This review outlines the historical development of therapeutic cancer vaccines, neoantigen identification strategies, and recent progress across DNA, RNA, peptide, cellular, and viral vaccine platforms, and discusses key mechanisms shaping vaccine response and resistance, including pattern-recognition receptor signaling, dendritic cell-mediated antigen presentation, T-cell effector and memory differentiation, metabolic adaptation, epitope spreading, and tumor microenvironment remodeling, proposing that future vaccines be developed as integrated immunological systems coordinating antigen discovery, precise delivery, innate immune calibration, memory maintenance, and local immune suppression reversal.
This review outlines the historical development of therapeutic cancer vaccines, neoantigen identification strategies, and recent progress across DNA, RNA, peptide, cellular, and viral vaccine platforms, and discusses key mechanisms shaping vaccine response and resistance, including pattern-recognition receptor signaling, dendritic cell-mediated antigen presentation, T-cell effector and memory differentiation, metabolic adaptation, epitope spreading, and tumor microenvironment remodeling, proposing that future vaccines be developed as integrated immunological systems coordinating antigen discovery, precise delivery, innate immune calibration, memory maintenance, and local immune suppression reversal.
This review outlines the historical development of therapeutic cancer vaccines, neoantigen identification strategies, and recent progress across DNA, RNA, peptide, cellular, and viral vaccine platforms, and discusses key mechanisms shaping vaccine response and resistance, including pattern-recognition receptor signaling, dendritic cell-mediated antigen presentation, T-cell effector and memory differentiation, metabolic adaptation, epitope spreading, and tumor microenvironment remodeling, proposing that future vaccines be developed as integrated immunological systems coordinating antigen discovery, precise delivery, innate immune calibration, memory maintenance, and local immune suppression reversal.
This review outlines the historical development of therapeutic cancer vaccines, neoantigen identification strategies, and recent progress across DNA, RNA, peptide, cellular, and viral vaccine platforms, and discusses key mechanisms shaping vaccine response and resistance, including pattern-recognition receptor signaling, dendritic cell-mediated antigen presentation, T-cell effector and memory differentiation, metabolic adaptation, epitope spreading, and tumor microenvironment remodeling, proposing that future vaccines be developed as integrated immunological systems coordinating antigen discovery, precise delivery, innate immune calibration, memory maintenance, and local immune suppression reversal.
Discover Nano This study builds Ti3C2Tx MXene/n-type monocrystalline silicon heterojunction solar cells, exposes them to 50 MeV protons and 1 MeV electrons (electron fluence 10^12–10^15 particles/cm^2, proton 10^12–10^14 particles/cm^2), characterizes them with XRD, Raman, SEM/TEM, EBSD, APT and LAMMPS atomistic simulation, and couples a physics-informed digital twin with Random Forest Regression, reporting over 84% retention of initial power conversion efficiency after irradiation versus about 55% for conventional Si cells, with R^2 > 0.96 for predictions of VOC, JSC, FF and PCE.
This study builds Ti3C2Tx MXene/n-type monocrystalline silicon heterojunction solar cells, exposes them to 50 MeV protons and 1 MeV electrons (electron fluence 10^12–10^15 particles/cm^2, proton 10^12–10^14 particles/cm^2), characterizes them with XRD, Raman, SEM/TEM, EBSD, APT and LAMMPS atomistic simulation, and couples a physics-informed digital twin with Random Forest Regression, reporting over 84% retention of initial power conversion efficiency after irradiation versus about 55% for conventional Si cells, with R^2 > 0.96 for predictions of VOC, JSC, FF and PCE.
This study builds Ti3C2Tx MXene/n-type monocrystalline silicon heterojunction solar cells, exposes them to 50 MeV protons and 1 MeV electrons (electron fluence 10^12–10^15 particles/cm^2, proton 10^12–10^14 particles/cm^2), characterizes them with XRD, Raman, SEM/TEM, EBSD, APT and LAMMPS atomistic simulation, and couples a physics-informed digital twin with Random Forest Regression, reporting over 84% retention of initial power conversion efficiency after irradiation versus about 55% for conventional Si cells, with R^2 > 0.96 for predictions of VOC, JSC, FF and PCE.
This study builds Ti3C2Tx MXene/n-type monocrystalline silicon heterojunction solar cells, exposes them to 50 MeV protons and 1 MeV electrons (electron fluence 10^12–10^15 particles/cm^2, proton 10^12–10^14 particles/cm^2), characterizes them with XRD, Raman, SEM/TEM, EBSD, APT and LAMMPS atomistic simulation, and couples a physics-informed digital twin with Random Forest Regression, reporting over 84% retention of initial power conversion efficiency after irradiation versus about 55% for conventional Si cells, with R^2 > 0.96 for predictions of VOC, JSC, FF and PCE.
npj Computational Materials This work introduces a modular, large language model (LLM)-powered agent pipeline that automatically extracts and structures composition–process–property attributes of concrete materials from tables and text in scientific publications, achieving F1 scores up to 0.98 across 17 open and proprietary models and, within about one hour, extracting over 10,000 records from 278 papers screened from more than 27,000 publications; after postprocessing this yields the largest open laboratory database for blended cement concrete with nearly 9,000 high-quality records and over 100 attributes, and machine learning analyses indicate that large, diverse, information-rich datasets improve both in-distribution accuracy and out-of-distribution generalization to unseen materials systems.
This work introduces a modular, large language model (LLM)-powered agent pipeline that automatically extracts and structures composition–process–property attributes of concrete materials from tables and text in scientific publications, achieving F1 scores up to 0.98 across 17 open and proprietary models and, within about one hour, extracting over 10,000 records from 278 papers screened from more than 27,000 publications; after postprocessing this yields the largest open laboratory database for blended cement concrete with nearly 9,000 high-quality records and over 100 attributes, and machine learning analyses indicate that large, diverse, information-rich datasets improve both in-distribution accuracy and out-of-distribution generalization to unseen materials systems.
This work introduces a modular, large language model (LLM)-powered agent pipeline that automatically extracts and structures composition–process–property attributes of concrete materials from tables and text in scientific publications, achieving F1 scores up to 0.98 across 17 open and proprietary models and, within about one hour, extracting over 10,000 records from 278 papers screened from more than 27,000 publications; after postprocessing this yields the largest open laboratory database for blended cement concrete with nearly 9,000 high-quality records and over 100 attributes, and machine learning analyses indicate that large, diverse, information-rich datasets improve both in-distribution accuracy and out-of-distribution generalization to unseen materials systems.
This work introduces a modular, large language model (LLM)-powered agent pipeline that automatically extracts and structures composition–process–property attributes of concrete materials from tables and text in scientific publications, achieving F1 scores up to 0.98 across 17 open and proprietary models and, within about one hour, extracting over 10,000 records from 278 papers screened from more than 27,000 publications; after postprocessing this yields the largest open laboratory database for blended cement concrete with nearly 9,000 high-quality records and over 100 attributes, and machine learning analyses indicate that large, diverse, information-rich datasets improve both in-distribution accuracy and out-of-distribution generalization to unseen materials systems.
Scientific Reports This study evaluated 11 vision models from the ResNet, DenseNet, and Vision Transformer families (seven trained from scratch, four pretrained) on the Oxford-IIIT Pet and Chest X-ray Pneumothorax datasets with ground-truth segmentation masks, generated explanations with five post-hoc XAI methods (Saliency, GradientSHAP, Integrated Gradients, Feature Permutation, Grad-CAM), and quantified mask alignment using Relevance Rank Accuracy and the proposed Dual-Polarity Precision; it found that increasing depth and parameter count did not improve explanation quality in most statistical comparisons, that smaller models often matched or exceeded deeper variants, that pretraining typically improved predictive performance and increased the dependence of explanations on learned weights without consisten
This study evaluated 11 vision models from the ResNet, DenseNet, and Vision Transformer families (seven trained from scratch, four pretrained) on the Oxford-IIIT Pet and Chest X-ray Pneumothorax datasets with ground-truth segmentation masks, generated explanations with five post-hoc XAI methods (Saliency, GradientSHAP, Integrated Gradients, Feature Permutation, Grad-CAM), and quantified mask alignment using Relevance Rank Accuracy and the proposed Dual-Polarity Precision; it found that increasing depth and parameter count did not improve explanation quality in most statistical comparisons, that smaller models often matched or exceeded deeper variants, that pretraining typically improved predictive performance and increased the dependence of explanations on learned weights without consisten
This study evaluated 11 vision models from the ResNet, DenseNet, and Vision Transformer families (seven trained from scratch, four pretrained) on the Oxford-IIIT Pet and Chest X-ray Pneumothorax datasets with ground-truth segmentation masks, generated explanations with five post-hoc XAI methods (Saliency, GradientSHAP, Integrated Gradients, Feature Permutation, Grad-CAM), and quantified mask alignment using Relevance Rank Accuracy and the proposed Dual-Polarity Precision; it found that increasing depth and parameter count did not improve explanation quality in most statistical comparisons, that smaller models often matched or exceeded deeper variants, that pretraining typically improved predictive performance and increased the dependence of explanations on learned weights without consisten
This study evaluated 11 vision models from the ResNet, DenseNet, and Vision Transformer families (seven trained from scratch, four pretrained) on the Oxford-IIIT Pet and Chest X-ray Pneumothorax datasets with ground-truth segmentation masks, generated explanations with five post-hoc XAI methods (Saliency, GradientSHAP, Integrated Gradients, Feature Permutation, Grad-CAM), and quantified mask alignment using Relevance Rank Accuracy and the proposed Dual-Polarity Precision; it found that increasing depth and parameter count did not improve explanation quality in most statistical comparisons, that smaller models often matched or exceeded deeper variants, that pretraining typically improved predictive performance and increased the dependence of explanations on learned weights without consisten
Health Science Reports Following PRISMA, this review searched PubMed, Web of Science, and Scopus (January 1, 2019 to March 27, 2025), included 58 of 1923 records, and used the CRISP-DM phases (problem understanding, data understanding, data preparation, modeling, evaluation, deployment) to organize machine learning approaches to noninvasive anemia detection and hemoglobin estimation, covering algorithms, data sources, acquisition sites, light sources, evaluation metrics, and deployment, alongside a PROBAST appraisal of risk of bias and applicability.
Following PRISMA, this review searched PubMed, Web of Science, and Scopus (January 1, 2019 to March 27, 2025), included 58 of 1923 records, and used the CRISP-DM phases (problem understanding, data understanding, data preparation, modeling, evaluation, deployment) to organize machine learning approaches to noninvasive anemia detection and hemoglobin estimation, covering algorithms, data sources, acquisition sites, light sources, evaluation metrics, and deployment, alongside a PROBAST appraisal of risk of bias and applicability.
Following PRISMA, this review searched PubMed, Web of Science, and Scopus (January 1, 2019 to March 27, 2025), included 58 of 1923 records, and used the CRISP-DM phases (problem understanding, data understanding, data preparation, modeling, evaluation, deployment) to organize machine learning approaches to noninvasive anemia detection and hemoglobin estimation, covering algorithms, data sources, acquisition sites, light sources, evaluation metrics, and deployment, alongside a PROBAST appraisal of risk of bias and applicability.
Following PRISMA, this review searched PubMed, Web of Science, and Scopus (January 1, 2019 to March 27, 2025), included 58 of 1923 records, and used the CRISP-DM phases (problem understanding, data understanding, data preparation, modeling, evaluation, deployment) to organize machine learning approaches to noninvasive anemia detection and hemoglobin estimation, covering algorithms, data sources, acquisition sites, light sources, evaluation metrics, and deployment, alongside a PROBAST appraisal of risk of bias and applicability.
medRxiv This study constructed a multilingual clinical dictation corpus (five clinical dictation scripts spanning a complexity gradient, translated into 99 languages, rendered to synthetic speech under three acoustic conditions, and transcribed by a production ambient scribe), computed six frequency metrics, and had three independent large language model raters assess clinically meaningful error patterns using a Severity x Likelihood framework; across 59,819 genuine transcription-error occurrences, 58,329 (97.5%) were LOW risk and 251 (0.42%) CRITICAL or HIGH, none of the six frequency metrics showed a detectable association with serious clinical risk (absolute Spearman rho < 0.16), a Severity x Likelihood sum remained strongly correlated with WER (rho=0.
This study constructed a multilingual clinical dictation corpus (five clinical dictation scripts spanning a complexity gradient, translated into 99 languages, rendered to synthetic speech under three acoustic conditions, and transcribed by a production ambient scribe), computed six frequency metrics, and had three independent large language model raters assess clinically meaningful error patterns using a Severity x Likelihood framework; across 59,819 genuine transcription-error occurrences, 58,329 (97.5%) were LOW risk and 251 (0.42%) CRITICAL or HIGH, none of the six frequency metrics showed a detectable association with serious clinical risk (absolute Spearman rho < 0.16), a Severity x Likelihood sum remained strongly correlated with WER (rho=0.
This study constructed a multilingual clinical dictation corpus (five clinical dictation scripts spanning a complexity gradient, translated into 99 languages, rendered to synthetic speech under three acoustic conditions, and transcribed by a production ambient scribe), computed six frequency metrics, and had three independent large language model raters assess clinically meaningful error patterns using a Severity x Likelihood framework; across 59,819 genuine transcription-error occurrences, 58,329 (97.5%) were LOW risk and 251 (0.42%) CRITICAL or HIGH, none of the six frequency metrics showed a detectable association with serious clinical risk (absolute Spearman rho < 0.16), a Severity x Likelihood sum remained strongly correlated with WER (rho=0.
This study constructed a multilingual clinical dictation corpus (five clinical dictation scripts spanning a complexity gradient, translated into 99 languages, rendered to synthetic speech under three acoustic conditions, and transcribed by a production ambient scribe), computed six frequency metrics, and had three independent large language model raters assess clinically meaningful error patterns using a Severity x Likelihood framework; across 59,819 genuine transcription-error occurrences, 58,329 (97.5%) were LOW risk and 251 (0.42%) CRITICAL or HIGH, none of the six frequency metrics showed a detectable association with serious clinical risk (absolute Spearman rho < 0.16), a Severity x Likelihood sum remained strongly correlated with WER (rho=0.
Knowledge-Based Systems This work proposes RIM (Retrieval-In-Matching), which renders UAV-viewpoint references from Google 3D Tiles across locations, altitudes, and orientations, adapts SALAD in two stages with pose-near positives and geographically distant hard negatives and then re-ranks Top-K candidates by local geometric consistency, and further freezes the adapted DINOv2-B retriever while distilling a local-descriptor decoder from its token field and a shallow VGG19 detail stream so that one query-side DINOv2-B forward supports both SALAD retrieval and local description; evaluated zero-shot on the reconstructed EPFL Urbanscape and self-collected Chang'an Park datasets, both geographically disjoint from the training data, RIM outperforms ten retrieval baselines, improving Recall@1 over SALAD by 8.55/13.
This work proposes RIM (Retrieval-In-Matching), which renders UAV-viewpoint references from Google 3D Tiles across locations, altitudes, and orientations, adapts SALAD in two stages with pose-near positives and geographically distant hard negatives and then re-ranks Top-K candidates by local geometric consistency, and further freezes the adapted DINOv2-B retriever while distilling a local-descriptor decoder from its token field and a shallow VGG19 detail stream so that one query-side DINOv2-B forward supports both SALAD retrieval and local description; evaluated zero-shot on the reconstructed EPFL Urbanscape and self-collected Chang'an Park datasets, both geographically disjoint from the training data, RIM outperforms ten retrieval baselines, improving Recall@1 over SALAD by 8.55/13.
This work proposes RIM (Retrieval-In-Matching), which renders UAV-viewpoint references from Google 3D Tiles across locations, altitudes, and orientations, adapts SALAD in two stages with pose-near positives and geographically distant hard negatives and then re-ranks Top-K candidates by local geometric consistency, and further freezes the adapted DINOv2-B retriever while distilling a local-descriptor decoder from its token field and a shallow VGG19 detail stream so that one query-side DINOv2-B forward supports both SALAD retrieval and local description; evaluated zero-shot on the reconstructed EPFL Urbanscape and self-collected Chang'an Park datasets, both geographically disjoint from the training data, RIM outperforms ten retrieval baselines, improving Recall@1 over SALAD by 8.55/13.
This work proposes RIM (Retrieval-In-Matching), which renders UAV-viewpoint references from Google 3D Tiles across locations, altitudes, and orientations, adapts SALAD in two stages with pose-near positives and geographically distant hard negatives and then re-ranks Top-K candidates by local geometric consistency, and further freezes the adapted DINOv2-B retriever while distilling a local-descriptor decoder from its token field and a shallow VGG19 detail stream so that one query-side DINOv2-B forward supports both SALAD retrieval and local description; evaluated zero-shot on the reconstructed EPFL Urbanscape and self-collected Chang'an Park datasets, both geographically disjoint from the training data, RIM outperforms ten retrieval baselines, improving Recall@1 over SALAD by 8.55/13.
medRxiv Using 5,613 autopsy cases from the Mayo Clinic Brain Bank collected between 1998 and 2023, this study fine-tuned a large language model to convert narrative gross descriptions into semi-quantitative scores for 39 features (extraction accuracy 0.95 on 200 manually annotated feature-level test examples), then classified seven neuropathologic diagnostic categories with a CatBoost classifier and a second fine-tuned LLM, both including age at death, sex, and brain weight; on a held-out test set of 562 cases CatBoost reached accuracy 0.73, kappa 0.65, and macro-average AUC 0.92, while the text-based LLM reached accuracy 0.75 and kappa 0.68, with macro-average sensitivity 0.66 for both, PSP sensitivity 0.92 and 0.93, MSA sensitivity 0.87 and 0.92, but AD-LBD sensitivity only 0.21 and 0.
Using 5,613 autopsy cases from the Mayo Clinic Brain Bank collected between 1998 and 2023, this study fine-tuned a large language model to convert narrative gross descriptions into semi-quantitative scores for 39 features (extraction accuracy 0.95 on 200 manually annotated feature-level test examples), then classified seven neuropathologic diagnostic categories with a CatBoost classifier and a second fine-tuned LLM, both including age at death, sex, and brain weight; on a held-out test set of 562 cases CatBoost reached accuracy 0.73, kappa 0.65, and macro-average AUC 0.92, while the text-based LLM reached accuracy 0.75 and kappa 0.68, with macro-average sensitivity 0.66 for both, PSP sensitivity 0.92 and 0.93, MSA sensitivity 0.87 and 0.92, but AD-LBD sensitivity only 0.21 and 0.
Using 5,613 autopsy cases from the Mayo Clinic Brain Bank collected between 1998 and 2023, this study fine-tuned a large language model to convert narrative gross descriptions into semi-quantitative scores for 39 features (extraction accuracy 0.95 on 200 manually annotated feature-level test examples), then classified seven neuropathologic diagnostic categories with a CatBoost classifier and a second fine-tuned LLM, both including age at death, sex, and brain weight; on a held-out test set of 562 cases CatBoost reached accuracy 0.73, kappa 0.65, and macro-average AUC 0.92, while the text-based LLM reached accuracy 0.75 and kappa 0.68, with macro-average sensitivity 0.66 for both, PSP sensitivity 0.92 and 0.93, MSA sensitivity 0.87 and 0.92, but AD-LBD sensitivity only 0.21 and 0.
Using 5,613 autopsy cases from the Mayo Clinic Brain Bank collected between 1998 and 2023, this study fine-tuned a large language model to convert narrative gross descriptions into semi-quantitative scores for 39 features (extraction accuracy 0.95 on 200 manually annotated feature-level test examples), then classified seven neuropathologic diagnostic categories with a CatBoost classifier and a second fine-tuned LLM, both including age at death, sex, and brain weight; on a held-out test set of 562 cases CatBoost reached accuracy 0.73, kappa 0.65, and macro-average AUC 0.92, while the text-based LLM reached accuracy 0.75 and kappa 0.68, with macro-average sensitivity 0.66 for both, PSP sensitivity 0.92 and 0.93, MSA sensitivity 0.87 and 0.92, but AD-LBD sensitivity only 0.21 and 0.
medRxiv This work presents ISDS-MRI, a unified framework that uses graph neural networks and sequence-specific feature modeling to detect ischemic stroke, segment lesions, and estimate lesion volume from multi-sequence MRI while accommodating incomplete sequence combinations; evaluated across multiple public MRI datasets and a newly curated BGD-MRIS dataset of 532 scans from three hospitals in Bangladesh, it achieves a Dice score of 0.725, an AUC of 0.962, and a lesion volume estimation relative error of 8.4%, outperforming comparison methods by 3.2% in Dice, 2.6% in detection, and reducing volume relative error by 1.9%.
This work presents ISDS-MRI, a unified framework that uses graph neural networks and sequence-specific feature modeling to detect ischemic stroke, segment lesions, and estimate lesion volume from multi-sequence MRI while accommodating incomplete sequence combinations; evaluated across multiple public MRI datasets and a newly curated BGD-MRIS dataset of 532 scans from three hospitals in Bangladesh, it achieves a Dice score of 0.725, an AUC of 0.962, and a lesion volume estimation relative error of 8.4%, outperforming comparison methods by 3.2% in Dice, 2.6% in detection, and reducing volume relative error by 1.9%.
This work presents ISDS-MRI, a unified framework that uses graph neural networks and sequence-specific feature modeling to detect ischemic stroke, segment lesions, and estimate lesion volume from multi-sequence MRI while accommodating incomplete sequence combinations; evaluated across multiple public MRI datasets and a newly curated BGD-MRIS dataset of 532 scans from three hospitals in Bangladesh, it achieves a Dice score of 0.725, an AUC of 0.962, and a lesion volume estimation relative error of 8.4%, outperforming comparison methods by 3.2% in Dice, 2.6% in detection, and reducing volume relative error by 1.9%.
This work presents ISDS-MRI, a unified framework that uses graph neural networks and sequence-specific feature modeling to detect ischemic stroke, segment lesions, and estimate lesion volume from multi-sequence MRI while accommodating incomplete sequence combinations; evaluated across multiple public MRI datasets and a newly curated BGD-MRIS dataset of 532 scans from three hospitals in Bangladesh, it achieves a Dice score of 0.725, an AUC of 0.962, and a lesion volume estimation relative error of 8.4%, outperforming comparison methods by 3.2% in Dice, 2.6% in detection, and reducing volume relative error by 1.9%.
medRxiv This study investigated AI acceptability among parents (first cohort n = 198; second cohort n = 79) and pediatric healthcare professionals (n = 33) across different disease, diagnosis, or treatment scenarios, finding that more liberal data privacy was associated with reduced willingness to use AI (p < .001), higher perceived disease severity was linked to higher willingness to use AI in the second parent group (beta = .10, p = .013), and when AI and clinician recommendations conflicted, parents were more likely to choose AI over clinician judgment in treatment compared to diagnosis scenarios, with healthcare professionals showing similar patterns but additionally weighting perceived disease severity.
This study investigated AI acceptability among parents (first cohort n = 198; second cohort n = 79) and pediatric healthcare professionals (n = 33) across different disease, diagnosis, or treatment scenarios, finding that more liberal data privacy was associated with reduced willingness to use AI (p < .001), higher perceived disease severity was linked to higher willingness to use AI in the second parent group (beta = .10, p = .013), and when AI and clinician recommendations conflicted, parents were more likely to choose AI over clinician judgment in treatment compared to diagnosis scenarios, with healthcare professionals showing similar patterns but additionally weighting perceived disease severity.
This study investigated AI acceptability among parents (first cohort n = 198; second cohort n = 79) and pediatric healthcare professionals (n = 33) across different disease, diagnosis, or treatment scenarios, finding that more liberal data privacy was associated with reduced willingness to use AI (p < .001), higher perceived disease severity was linked to higher willingness to use AI in the second parent group (beta = .10, p = .013), and when AI and clinician recommendations conflicted, parents were more likely to choose AI over clinician judgment in treatment compared to diagnosis scenarios, with healthcare professionals showing similar patterns but additionally weighting perceived disease severity.
This study investigated AI acceptability among parents (first cohort n = 198; second cohort n = 79) and pediatric healthcare professionals (n = 33) across different disease, diagnosis, or treatment scenarios, finding that more liberal data privacy was associated with reduced willingness to use AI (p < .001), higher perceived disease severity was linked to higher willingness to use AI in the second parent group (beta = .10, p = .013), and when AI and clinician recommendations conflicted, parents were more likely to choose AI over clinician judgment in treatment compared to diagnosis scenarios, with healthcare professionals showing similar patterns but additionally weighting perceived disease severity.
Nature An AI Structural Biology (AISB) network of pharmaceutical companies fine-tuned the open-source model OpenFold3 on 20,167 proprietary protein-ligand structures from five companies; on a held-out test set of 1,056 protein-ligand structures, the model achieved high-accuracy predictions for more than half, compared with one-third for the public version of OpenFold3 and around 40% for the competing open-source model Boltz-2, and it also outperformed models trained only on individual companies' siloed data, indicating that pooling private data markedly improves protein-ligand interaction prediction.
An AI Structural Biology (AISB) network of pharmaceutical companies fine-tuned the open-source model OpenFold3 on 20,167 proprietary protein-ligand structures from five companies; on a held-out test set of 1,056 protein-ligand structures, the model achieved high-accuracy predictions for more than half, compared with one-third for the public version of OpenFold3 and around 40% for the competing open-source model Boltz-2, and it also outperformed models trained only on individual companies' siloed data, indicating that pooling private data markedly improves protein-ligand interaction prediction.
An AI Structural Biology (AISB) network of pharmaceutical companies fine-tuned the open-source model OpenFold3 on 20,167 proprietary protein-ligand structures from five companies; on a held-out test set of 1,056 protein-ligand structures, the model achieved high-accuracy predictions for more than half, compared with one-third for the public version of OpenFold3 and around 40% for the competing open-source model Boltz-2, and it also outperformed models trained only on individual companies' siloed data, indicating that pooling private data markedly improves protein-ligand interaction prediction.
An AI Structural Biology (AISB) network of pharmaceutical companies fine-tuned the open-source model OpenFold3 on 20,167 proprietary protein-ligand structures from five companies; on a held-out test set of 1,056 protein-ligand structures, the model achieved high-accuracy predictions for more than half, compared with one-third for the public version of OpenFold3 and around 40% for the competing open-source model Boltz-2, and it also outperformed models trained only on individual companies' siloed data, indicating that pooling private data markedly improves protein-ligand interaction prediction.
Journal of paediatrics and child health Using the search terms 'management of lower limb pain in children with hypermobility-associated conditions', this study collated webpages from Google searches and the corresponding overviews presented by Gemini, assessed webpage quality with the Health Information Website Evaluation Tool and AI overview quality with the Quality Assessment of Medical Artificial Intelligence, and evaluating 20 webpages and four AI-generated overviews found that webpage quality was mostly moderate (14 moderate, three good) while all AI overviews were rated good, with accuracy the lowest-scoring domain for webpages and completeness and provision of resources and references the lowest-scoring domains for AI overviews.
Using the search terms 'management of lower limb pain in children with hypermobility-associated conditions', this study collated webpages from Google searches and the corresponding overviews presented by Gemini, assessed webpage quality with the Health Information Website Evaluation Tool and AI overview quality with the Quality Assessment of Medical Artificial Intelligence, and evaluating 20 webpages and four AI-generated overviews found that webpage quality was mostly moderate (14 moderate, three good) while all AI overviews were rated good, with accuracy the lowest-scoring domain for webpages and completeness and provision of resources and references the lowest-scoring domains for AI overviews.
Using the search terms 'management of lower limb pain in children with hypermobility-associated conditions', this study collated webpages from Google searches and the corresponding overviews presented by Gemini, assessed webpage quality with the Health Information Website Evaluation Tool and AI overview quality with the Quality Assessment of Medical Artificial Intelligence, and evaluating 20 webpages and four AI-generated overviews found that webpage quality was mostly moderate (14 moderate, three good) while all AI overviews were rated good, with accuracy the lowest-scoring domain for webpages and completeness and provision of resources and references the lowest-scoring domains for AI overviews.
Using the search terms 'management of lower limb pain in children with hypermobility-associated conditions', this study collated webpages from Google searches and the corresponding overviews presented by Gemini, assessed webpage quality with the Health Information Website Evaluation Tool and AI overview quality with the Quality Assessment of Medical Artificial Intelligence, and evaluating 20 webpages and four AI-generated overviews found that webpage quality was mostly moderate (14 moderate, three good) while all AI overviews were rated good, with accuracy the lowest-scoring domain for webpages and completeness and provision of resources and references the lowest-scoring domains for AI overviews.
Terence Tao blog RSS This guest post by Bryna Kra uses the Nivat conjecture to argue that AI has sharply lowered the cost of producing sophisticated proofs, as shown by several purported proofs she received this week, and that because a proof is more than a certificate of correctness—it is understanding, explanation, and collective knowledge—the mathematical community must redefine and reward discovery, proof, formalization, and exposition.
This guest post by Bryna Kra uses the Nivat conjecture to argue that AI has sharply lowered the cost of producing sophisticated proofs, as shown by several purported proofs she received this week, and that because a proof is more than a certificate of correctness—it is understanding, explanation, and collective knowledge—the mathematical community must redefine and reward discovery, proof, formalization, and exposition.
This guest post by Bryna Kra uses the Nivat conjecture to argue that AI has sharply lowered the cost of producing sophisticated proofs, as shown by several purported proofs she received this week, and that because a proof is more than a certificate of correctness—it is understanding, explanation, and collective knowledge—the mathematical community must redefine and reward discovery, proof, formalization, and exposition.
This guest post by Bryna Kra uses the Nivat conjecture to argue that AI has sharply lowered the cost of producing sophisticated proofs, as shown by several purported proofs she received this week, and that because a proof is more than a certificate of correctness—it is understanding, explanation, and collective knowledge—the mathematical community must redefine and reward discovery, proof, formalization, and exposition.
Terence Tao blog RSS This essay by Nestor Guillen argues that large language models (LLMs) should be understood as a kind of cultural and social technology, like markets, bureaucracies, and the scientific literature, which aggregates and compresses human accumulated information, so that when an LLM output contains a new mathematical idea it should be received as the fruit of mathematics' shared heritage rather than as a defeat for mathematicians; drawing on Farrell, Gopnik, Shalizi, and Evans's claim that large models are cultural technologies, the author proposes the metaphor of a 'convex hull of ideas,' suggesting that whether an LLM can solve a given mathematical problem depends largely on the state of the mathematical literature at the time rather than merely on model scale, illustrating this with the Kryl
This essay by Nestor Guillen argues that large language models (LLMs) should be understood as a kind of cultural and social technology, like markets, bureaucracies, and the scientific literature, which aggregates and compresses human accumulated information, so that when an LLM output contains a new mathematical idea it should be received as the fruit of mathematics' shared heritage rather than as a defeat for mathematicians; drawing on Farrell, Gopnik, Shalizi, and Evans's claim that large models are cultural technologies, the author proposes the metaphor of a 'convex hull of ideas,' suggesting that whether an LLM can solve a given mathematical problem depends largely on the state of the mathematical literature at the time rather than merely on model scale, illustrating this with the Kryl
This essay by Nestor Guillen argues that large language models (LLMs) should be understood as a kind of cultural and social technology, like markets, bureaucracies, and the scientific literature, which aggregates and compresses human accumulated information, so that when an LLM output contains a new mathematical idea it should be received as the fruit of mathematics' shared heritage rather than as a defeat for mathematicians; drawing on Farrell, Gopnik, Shalizi, and Evans's claim that large models are cultural technologies, the author proposes the metaphor of a 'convex hull of ideas,' suggesting that whether an LLM can solve a given mathematical problem depends largely on the state of the mathematical literature at the time rather than merely on model scale, illustrating this with the Kryl
This essay by Nestor Guillen argues that large language models (LLMs) should be understood as a kind of cultural and social technology, like markets, bureaucracies, and the scientific literature, which aggregates and compresses human accumulated information, so that when an LLM output contains a new mathematical idea it should be received as the fruit of mathematics' shared heritage rather than as a defeat for mathematicians; drawing on Farrell, Gopnik, Shalizi, and Evans's claim that large models are cultural technologies, the author proposes the metaphor of a 'convex hull of ideas,' suggesting that whether an LLM can solve a given mathematical problem depends largely on the state of the mathematical literature at the time rather than merely on model scale, illustrating this with the Kryl
Nature News A consortium of pharmaceutical companies called the AI Structural Biology (AISB) Network fine-tuned OpenFold3, previously trained only on PDB data, on 20,167 protein–ligand structures from five companies, and on a held-out test set of 1,056 protein–ligand structures the model reached high accuracy on more than half of them, versus about one-third for the public OpenFold3 and around 40% for the open-source model Boltz-2, while also outperforming models trained only on each company's own data, indicating that pooling data is more valuable than keeping it siloed.
A consortium of pharmaceutical companies called the AI Structural Biology (AISB) Network fine-tuned OpenFold3, previously trained only on PDB data, on 20,167 protein–ligand structures from five companies, and on a held-out test set of 1,056 protein–ligand structures the model reached high accuracy on more than half of them, versus about one-third for the public OpenFold3 and around 40% for the open-source model Boltz-2, while also outperforming models trained only on each company's own data, indicating that pooling data is more valuable than keeping it siloed.
A consortium of pharmaceutical companies called the AI Structural Biology (AISB) Network fine-tuned OpenFold3, previously trained only on PDB data, on 20,167 protein–ligand structures from five companies, and on a held-out test set of 1,056 protein–ligand structures the model reached high accuracy on more than half of them, versus about one-third for the public OpenFold3 and around 40% for the open-source model Boltz-2, while also outperforming models trained only on each company's own data, indicating that pooling data is more valuable than keeping it siloed.
A consortium of pharmaceutical companies called the AI Structural Biology (AISB) Network fine-tuned OpenFold3, previously trained only on PDB data, on 20,167 protein–ligand structures from five companies, and on a held-out test set of 1,056 protein–ligand structures the model reached high accuracy on more than half of them, versus about one-third for the public OpenFold3 and around 40% for the open-source model Boltz-2, while also outperforming models trained only on each company's own data, indicating that pooling data is more valuable than keeping it siloed.
Nature Medicine Whole-genome sequencing of 7,851 parent–offspring families identified de novo mutations with parent-of-origin and post-zygotic features that were associated with distinct assisted reproductive technology (ART) procedures independently of parental age at conception, and increased paternal mutational burden statistically mediated the effects of advanced parental age and ART on gestational duration and other birth outcomes.
Whole-genome sequencing of 7,851 parent–offspring families identified de novo mutations with parent-of-origin and post-zygotic features that were associated with distinct assisted reproductive technology (ART) procedures independently of parental age at conception, and increased paternal mutational burden statistically mediated the effects of advanced parental age and ART on gestational duration and other birth outcomes.
Whole-genome sequencing of 7,851 parent–offspring families identified de novo mutations with parent-of-origin and post-zygotic features that were associated with distinct assisted reproductive technology (ART) procedures independently of parental age at conception, and increased paternal mutational burden statistically mediated the effects of advanced parental age and ART on gestational duration and other birth outcomes.
Whole-genome sequencing of 7,851 parent–offspring families identified de novo mutations with parent-of-origin and post-zygotic features that were associated with distinct assisted reproductive technology (ART) procedures independently of parental age at conception, and increased paternal mutational burden statistically mediated the effects of advanced parental age and ART on gestational duration and other birth outcomes.
Nature News This Nature news report describes the rise of science philanthropy in Brazil: over the past decade the country saw the founding of its first non-profit dedicated to funding scientific research, the Serrapilheira Institute, which has distributed more than US$25 million to early-career researchers, alongside other funders such as Pioneer Science and the Pensi Institute, a 2023 research centre at the University of São Paulo to study and foster science philanthropy, and early efforts to form a local association inspired by the US Science Philanthropy Alliance.
This Nature news report describes the rise of science philanthropy in Brazil: over the past decade the country saw the founding of its first non-profit dedicated to funding scientific research, the Serrapilheira Institute, which has distributed more than US$25 million to early-career researchers, alongside other funders such as Pioneer Science and the Pensi Institute, a 2023 research centre at the University of São Paulo to study and foster science philanthropy, and early efforts to form a local association inspired by the US Science Philanthropy Alliance.
This Nature news report describes the rise of science philanthropy in Brazil: over the past decade the country saw the founding of its first non-profit dedicated to funding scientific research, the Serrapilheira Institute, which has distributed more than US$25 million to early-career researchers, alongside other funders such as Pioneer Science and the Pensi Institute, a 2023 research centre at the University of São Paulo to study and foster science philanthropy, and early efforts to form a local association inspired by the US Science Philanthropy Alliance.
This Nature news report describes the rise of science philanthropy in Brazil: over the past decade the country saw the founding of its first non-profit dedicated to funding scientific research, the Serrapilheira Institute, which has distributed more than US$25 million to early-career researchers, alongside other funders such as Pioneer Science and the Pensi Institute, a 2023 research centre at the University of São Paulo to study and foster science philanthropy, and early efforts to form a local association inspired by the US Science Philanthropy Alliance.
Nature News A proof-of-concept study reports that a single implanted array of 253 electrodes on the brain's sensorimotor cortex can capture neural signals related to both speech and body movement and use artificial intelligence to turn brain activity into on-screen text while driving a personalized animated avatar, decoding verbal and non-verbal communication simultaneously within seconds of the user's intent.
A proof-of-concept study reports that a single implanted array of 253 electrodes on the brain's sensorimotor cortex can capture neural signals related to both speech and body movement and use artificial intelligence to turn brain activity into on-screen text while driving a personalized animated avatar, decoding verbal and non-verbal communication simultaneously within seconds of the user's intent.
A proof-of-concept study reports that a single implanted array of 253 electrodes on the brain's sensorimotor cortex can capture neural signals related to both speech and body movement and use artificial intelligence to turn brain activity into on-screen text while driving a personalized animated avatar, decoding verbal and non-verbal communication simultaneously within seconds of the user's intent.
A proof-of-concept study reports that a single implanted array of 253 electrodes on the brain's sensorimotor cortex can capture neural signals related to both speech and body movement and use artificial intelligence to turn brain activity into on-screen text while driving a personalized animated avatar, decoding verbal and non-verbal communication simultaneously within seconds of the user's intent.