Supplementary data: Interpretable Machine Learning for the Rational Design and In Vitro Validation of SET-M33 Dendrimeric Antimicrobial Peptides

This repository contains the data, models, and analysis files supporting the study "Interpretable Machine Learning for the Rational Design and In Vitro Validation of SET-M33 Dendrimeric Antimicrobial Peptides". The work develops an interpretable machine-learning framework to design and prioritize species-specific SET-M33 antimicrobial peptide analogs against Escherichia coli, Pseudomonas aeruginosa, and Klebsiella pneumoniae, and validates selected candidates in vitro. The pipeline comprises: (1) Random Forest MIC-prediction models trained on species-specific physicochemical descriptor pools (StarPep/ProtDCal descriptors, WEKA feature selection); (2) generation and prediction of N-terminally elongated cationic analogs; (3) organization of analogs into High-Similarity Physicochemical Networks (HSPN) with Louvain modules; (4) interpretation of activity drivers through SHAP and UMAP analysis; and (5) selection and in vitro MIC validation of candidates. The repository is organized into Training (databases, descriptor pools, and trained models), Test (generated analogs and predictions), Similarity networks (threshold sweeps and selected networks), Post-analysis (SHAP and UMAP outputs), and MIC test (in vitro results for synthesized candidates). A Code folder with the analysis notebooks (HSPN construction, candidate selection, SHAP, and functional-space analysis), documented in an accompanying README, is also included. A README file describing the full directory structure and naming conventions is included.本仓库包含支持研究《可解释机器学习用于SET-M33树状抗菌肽的理性设计与体外验证》的数据、模型与分析文件。 本研究构建了可解释机器学习框架，用于设计并筛选针对大肠杆菌（Escherichia coli）、铜绿假单胞菌（Pseudomonas aeruginosa）及肺炎克雷伯菌（Klebsiella pneumoniae）的物种特异性SET-M33抗菌肽类似物，并通过体外实验验证了筛选出的候选化合物。本研究的分析流程包含以下步骤：(1) 基于物种特异性理化描述符集（StarPep/ProtDCal描述符，采用WEKA进行特征选择）训练随机森林最小抑菌浓度（Minimum Inhibitory Concentration, MIC）预测模型；(2) 生成N端延长的阳离子类似物并开展活性预测；(3) 采用卢万（Louvain）模块将类似物构建为高相似度理化网络（High-Similarity Physicochemical Networks, HSPN）；(4) 通过SHapley可加解释（SHapley Additive exPlanations, SHAP）与均匀流形近似与投影（Uniform Manifold Approximation and Projection, UMAP）分析解析活性驱动因子；(5) 筛选候选化合物并开展体外MIC验证。 本仓库分为以下子模块：训练（Training，包含数据库、描述符集与已训练模型）、测试（Test，包含生成的类似物与预测结果）、相似度网络（Similarity networks，包含阈值扫描与筛选后的网络）、后分析（Post-analysis，包含SHAP与UMAP分析输出结果）以及MIC实验（MIC test，包含合成候选化合物的体外实验结果）。此外还设有Code文件夹，其中包含分析脚本笔记（涵盖高相似度理化网络构建、候选化合物筛选、SHAP分析与功能空间分析），并配有配套说明文档。本仓库同时包含一份README文件，用于完整说明整个目录结构与命名规范。