Agent Skills
› synthetic-sciences/openscience
› admet-reasoning
admet-reasoning
GitHub基于CoTox和DrugR的ADMET可解释性分析，将毒性风险映射至结构成因与生物通路，提供分子优化建议。用于药物化学团队的毒理评估及先导化合物优化规划。
Trigger Scenarios
需要解释ADMET预测结果中的具体毒性原因
制定分子结构优化策略以规避特定毒性
Install
npx skills add synthetic-sciences/openscience --skill admet-reasoning -g -y
SKILL.md
Frontmatter
{
"name": "admet-reasoning",
"tags": [
"drug-discovery",
"ADMET",
"toxicity",
"interpretation",
"reasoning"
],
"license": "MIT",
"version": "1.0.0",
"category": "chemistry",
"metadata": {
"skill-author": "Synthetic Sciences"
},
"description": "Interpretable ADMET analysis with mechanistic reasoning. Maps liabilities to structural causes and biological pathways. Based on CoTox (Park 2025) and DrugR (Liu 2026).",
"dependencies": [
"rdkit-pypi",
"numpy",
"pandas"
]
}
ADMET Reasoning
Overview
Standard ADMET prediction tools output scores (e.g., "hERG = 0.85") without explaining why. This skill adds mechanistic reasoning — mapping each ADMET liability to its structural cause, the biological mechanism it affects, and a suggested structural fix.
Based on:
- CoTox (Park et al., 2025): Chain-of-thought toxicity reasoning with structural + biological context improved F1 from 0.37 to 0.66
- DrugR (Liu et al., 2026): Explicit liability reasoning before optimization improved scores 18×
When to Use This Skill
- After ADMET prediction: Interpret flagged liabilities with structural explanations
- Lead optimization planning: Understand which structural features to modify and why
- Toxicity reports: Generate interpretable toxicity assessments for medicinal chemistry teams
- Design review: Evaluate whether proposed modifications address the right liabilities
Do NOT use this skill for:
- Raw ADMET score computation (use
admet-predictioninstead) - Molecular optimization (use
molecular-optimizationinstead)
Related Skills
- admet-prediction: Compute ADMET properties (run this first)
- molecular-optimization: Iterative optimization using liability analysis
- rdkit: Core molecular operations
Installation
Required dependencies
pip install rdkit-pypi numpy pandas
Core Workflows
1. Full Liability Report
Generate interpretable ADMET analysis for a molecule:
python scripts/reason_admet.py --smiles "c1ccc(NC(=O)c2ccccc2Cl)cc1" --output report.json
2. Batch Liability Analysis
python scripts/reason_admet.py --input compounds.csv --output liability_report.csv
3. Targeted Toxicity Reasoning
Focus on specific endpoints:
python scripts/reason_admet.py --smiles "CCN1CCCC1" --endpoints hERG,DILI,CYP --output tox_report.json
Script Reference
| Script | Purpose | Key Outputs |
|---|---|---|
reason_admet.py |
Full ADMET reasoning with structural explanations | JSON report with liabilities, causes, mechanisms, fixes |
Version History
- e9844a4 Current 2026-07-11 17:21
Dependencies
-
required
rdkit-pypi -
required
numpy - required Pandas分组操作保留原始多层索引
