Research map / Large language models / Agents, code and math
DrugTrail: Explainable Drug Discovery via Structured Reasoning and Druggability‑Tailored Preference Optimization
Yurou Liu, Mingyang Li, Xinyuan Zhu, Rui Jiao, Yiming Dong, Xinyu Tang, Yang Liu, Jieping Ye, Bing Su, Zheng Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add CCIR structured reasoning SFT stage (ablation) | Qwen3-1.7B · CrossDocked2020 | Vina docking energy: 9.77 -> -3.06 | Improved |
| add DTPO preference optimization on top of CCIR (ablation) | Qwen3-1.7B · CrossDocked2020 | IMP (%): 8.59 -> 33.48 | Improved |
| add DTPO on top of CCIR (ablation) | Qwen3-8B · CrossDocked2020 | Vina docking energy: -3.10 -> -6.82 | Improved |
| add DTPO on top of CCIR (ablation) | Qwen3-8B · CrossDocked2020 | QED: 0.31 -> 0.57 | Improved |
| remove CCIR stage and apply DTPO directly (ablation) | DrugTrail · CrossDocked2020 | Vina score and QED | Hurt |
| add reasoning quality reward (ablation) | DrugTrail GRPO · CrossDocked2020 | all aspects | Improved |
| add ligand-based reward (ablation) | DrugTrail GRPO · CrossDocked2020 | IMP and Vina score | Improved |
| add rule-based reward (ablation) | DrugTrail GRPO · CrossDocked2020 | LPSK | Improved |
| increase rule-based reward weight (ablation) | DrugTrail GRPO · CrossDocked2020 | QED vs binding | Mixed |
| add geometry-related reasoning dimensions (ablation) | DrugTrail · CrossDocked2020 | Vina score: -4.61 -> -6.79 | Improved |
| rank preference data by Boltz2 or SIU labels instead of Vina (ablation) | DrugTrail · CrossDocked2020 | Vina score: -6.79 vs -6.71 / -6.52 | No clear effect |
| apply DrugTrail zero-shot to protein GFP optimization (ablation) | Qwen3-8B · GFP medium difficulty | fitness: 0.08 -> 0.62 | Improved |
Closest papers by meaning
- Generating Novel Leads for Drug Discovery Using LLMs with Logical Feedback (AAAI 2024 · similarity 0.88)
- Knowledge Enhanced Representation Learning for Drug Discovery (AAAI 2024 · similarity 0.88)
- Beyond Chemical QA: Evaluating LLM's Chemical Reasoning with Modular Chemical Operations (NeurIPS 2025 · similarity 0.87)
- Drug Discovery with Dynamic Goal-aware Fragments (ICML 2024 · similarity 0.87)
- Efficient Evolutionary Search Over Chemical Space with Large Language Models (ICLR 2025 · similarity 0.86)
- A Dataset for Distilling Knowledge Priors from Literature for Therapeutic Design (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
