ChemLint — 一个MCP服务器，为LLM提供对化学信息学和分子机器学习工具的原生访问

---

摘要

ChemLint 将 AI 助手转变为强大的分子机器学习工作台。通过模型上下文协议（MCP），LLM 能够通过自然对话操作分子结构、计算描述符、训练机器学习模型并生成全面的分析报告。

只需与Claude Desktop（或任何MCP客户端）聊天，就能完成复杂的化学信息学工作流程，这些通常需要编写Python脚本并同时处理多个库。

---

✨ 主要特色

🧪 分子操作

• SMILES 处理：标准化、规范化、验证、10 步清理流程
• 分子描述符：简单（MW、LogP、TPSA）和复杂（ECFP、MACCS、RDKit指纹）
• 支架分析：Bemis-Murcko、通用支架、带有多样性度量的循环骨架
• 相似性与聚类：Tanimoto相似度、DBSCAN、层级、k-均值、Butina 聚类
• 亚结构匹配：SMARTS模式检测，内置88+功能基团

🤖 机器学习

🧬 ChemLint

ChemLint — An MCP server that gives LLMs native access to cheminformatics and molecular ML tools

---

Summary

ChemLint transforms AI assistants into powerful molecular machine learning workbenches. Through the Model Context Protocol (MCP), LLMs gain the ability to manipulate molecular structures, calculate descriptors, train ML models, and generate comprehensive analysis reports — all through natural conversation.

Simply chat with Claude Desktop (or any MCP client) to perform complex cheminformatics workflows that would normally require writing Python scripts and juggling multiple libraries.

---

✨ Key Features

🧪 Molecular Operations

• SMILES Processing: Standardization, canonicalization, validation, 10-step cleaning pipelines
• Molecular Descriptors: Simple (MW, LogP, TPSA) and complex (ECFP, MACCS, RDKit fingerprints)
• Scaffold Analysis: Bemis-Murcko, generic scaffolds, cyclic skeletons with diversity metrics
• Similarity & Clustering: Tanimoto similarity, DBSCAN, hierarchical, k-means, Butina clustering
• Substructure Matching: SMARTS pattern detection with 88+ built-in functional groups

🤖 Machine Learning

• 33 ML Algorithms: Classification & regression (RF, GBM, SVM, linear models, ensembles with uncertainty)
• Cross-Validation: 6 strategies (k-fold, stratified, Monte Carlo, scaffold, cluster, leave-P-out)
• Hyperparameter Tuning: Grid search, random search with customizable parameter spaces
• Model Evaluation: 20+ metrics, confusion matrices, ROC curves, calibration plots

📊 Quality Reports

• Data Quality Analysis: 19-section comprehensive report (PAINS, Lipinski, duplicates, stereochemistry, etc.)
• Split Quality Analysis: 8 data leakage checks (duplicates, similarity, scaffolds, stereoisomers)
• Scaffold Reports: Diversity metrics (Gini, Shannon entropy), enrichment analysis, structural outliers

📈 Visualization & Statistics

• Interactive Plots: Scatter plots with molecular tooltips, histograms, density plots, box plots, heatmaps
• Statistical Tests: 15+ tests (t-test, ANOVA, Mann-Whitney, Kruskal-Wallis, chi-square, normality tests)
• Dimensionality Reduction: PCA, t-SNE for chemical space visualization
• Outlier Detection: Z-score, IQR, isolation forest, local outlier factor

🔬 Advanced Features

• Activity Cliff Detection: Find structurally similar molecules with large activity differences
• Data Splitting: Random, stratified, scaffold-based, cluster-based, temporal splits
• Duplicate Handling: Activity conflict detection with aggregation strategies