CYPMol: A Single Model Framework Integrating Functional Residues with Protein Features and Molecule Embeddings to Predict CYP Substrates, Inhibitors, and Metabolism Sites.

Cytochrome P450 enzymes (CYPs) mediate xenobiotic metabolism in humans, and models for predicting CYP-molecule interactions and reactions, including substrates, inhibitors, and metabolism sites, are valuable tools for drug development. However, current prediction models typically rely solely on full-length CYP sequences, overlooking such functional residues that govern substrate binding and catalytic activity. Here, we present CYPMol, a deep learning framework for CYP substrates, inhibitors, and bonds of metabolism (BoMs) prediction tasks integrating functional residue and protein language features with pretrained small molecule embeddings in a single model architecture. Comparative analyses revealed that CYPMol could outperform current state-of-the-art models in predicting substrates (MCC = 0.819) and inhibitors (MCC = 0.725) of the nine human CYPs. CYPMol could also predict specific BoMs altered during catalysis for 537 animal, plant, or microbial CYPs. Our framework and accompanying data sets are publicly accessible at https://github.com/CjmTH/CYPMol or web server https://tianlab-tsinghua.cn/cypmol/ By incorporating functional residue information relevant to each task with other protein features, CYPMol provides a powerful framework for modeling CYP activity and interactions with small molecules, supporting both protein engineering and drug development.