Skip to main content

Matter Sciences

52 items

Journal of Chemical Information and Modeling Sep 14, 2026
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.
When Do Simple Models Win? Machine Learning Architectures for UV Absorption Prediction
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.
Discover Nano Sep 14, 2026
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.
Predictive Modelling and Experimental Analysis of Radiation-Resistant MXene–Silicon Heterojunction Solar Cells
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.

All results are shown

Page 6 · showing 2