MEN: leveraging explainable multimodal encoding network for precision prediction of CYP450 inhibitors

Area of Science:

• Computational chemistry and cheminformatics
• Pharmacology and drug discovery
• Artificial intelligence in medicine

Background:

• Drug-drug interactions (DDIs) pose significant clinical risks, often stemming from cytochrome P450 (CYP450) enzyme inhibition.
• Accurate prediction of CYP450 inhibitors is crucial for safe drug development, but current machine learning methods lack precision and interpretability.
• Existing approaches struggle to fully leverage diverse molecular data for robust CYP450 inhibition prediction.

Purpose of the Study:

• To develop an advanced machine learning model, the Multimodal Encoder Network (MEN), for enhanced prediction of CYP450 enzyme inhibitors.
• To improve the accuracy and biological interpretability of CYP450 inhibitor predictions by integrating multiple data modalities.
• To provide a tool that aids in identifying potential drug-drug interactions early in the drug development pipeline.

Main Methods:

• Developed a Multimodal Encoder Network (MEN) integrating three data types: chemical fingerprints (FEN), molecular graphs (GEN), and protein sequences (PEN).
• Employed specialized encoders for each data modality to extract complementary features, fused for a comprehensive representation.
• Incorporated an explainable AI (XAI) module with visualization techniques for biological interpretation of predictions.

Main Results:

• MEN achieved a high average accuracy of 93.7% across five CYP450 isoforms (1A2, 2C9, 2C19, 2D6, 3A4).
• Individual encoders demonstrated strong performance: FEN (80.8%), GEN (82.3%), and PEN (81.5%).
• Achieved excellent overall performance metrics, including AUC (98.5%), sensitivity (95.9%), specificity (97.2%), and MCC (88.2%).

Conclusions:

• The MEN model significantly advances the prediction of CYP450 inhibitors by effectively integrating multimodal molecular data.
• The integrated XAI module enhances biological interpretability, facilitating a deeper understanding of inhibition mechanisms.
• MEN offers a promising tool for improving drug safety and efficiency in pharmaceutical research and development.