Skip to main content
Search

# Retrosynthesis Planning

2 related articles

Subscribe to updates
  1. aclanthology.org

    Science rules move from the prompt into decoding: layered constraints lift average accuracy 12% across three scientific tasks

    Large language models still fabricate reasoning that contradicts domain rules in specialized scientific tasks. SciDC has a general model compile natural-language scientific knowledge into three layers of rules, top-level reasoning steps, middle-level conditional checks and bottom-level token masks, which a locally deployed domain model enforces during decoding. Across industrial formulation design, thyroid cancer TNM staging and single-step retrosynthesis, average accuracy rises 12% over vanilla generation, and ablations show removing the top or bottom layer sharply lowers formulation accuracy.

    Nanjing UniversityTsinghua University
    SciDCConstrained DecodingIndustrial Formulation DesignClinical Tumor DiagnosisRetrosynthesis Planning
  2. bioRxiv

    TRACEDD: Tool-grounded Reasoning and Agentic Coordination for Explainable Drug Design

    This work introduces TRACEDD, a tool-first multi-agent framework in which large language models serve as the reasoning and orchestration layer while domain-validated computational tools supply structural, chemical, pharmacological, and synthetic predictions, and it demonstrates an end-to-end JAK2 workflow spanning target validation, structure retrieval or AlphaFold prediction, druggable pocket identification, reinforcement-learning-based de novo molecular generation, lead optimization, ADMET and bioactivity evaluation, literature evidence integration, and retrosynthesis planning, with each decision linked to explicit tool invocations and intermediate evidence.

    Sarveswara Rao Vangala（First author）Arijit Roy（Corresponding author）
    TRACEDDTool-first orchestrationDe novo drug designRetrosynthesis PlanningDe Novo Molecular Generation

You’re all caught up