Residue-Level Feature Extraction from Protein–Peptide Complexes for Machine Learning Applications

Abstract In order to apply machine learning approaches for protein–peptide interaction analysis, it requires structured and quantitative features derived from three-dimensional molecular data. However, many existing workflows either rely on external tools or lack modular, lightweight implementations suitable for rapid feature engineering. Here, in this study I present a Python-based pipeline for residue-level feature extraction from protein–peptide complexes using Protein Data Bank (PDB) structures. The pipeline parses atomic coordinates using Biopython and computes a set of geometric, physicochemical, and interface-related descriptor. This descriptor includes distance-based metrics, contact counts, residue properties and flexibility proxies. The system is designed as a modular, reproducible pipeline that produces machine learning feature tables automatically. This approach provides a foundation for downstream predictive modeling tasks such as interface classification and binding prediction. The source code for the feature extraction pipeline is publicly available at:https://github.com/TaufiaHussain/cheminformatics-pdb-features. Example input structures were obtained from the Protein Data Bank (PDB), including the protein–peptide complex with PDB ID: 2EGN. The generated feature tables and example outputs are included in the repository.