Prediction of CYP450 Enzyme-Substrate Selectivity Based on the Network-Based Label Space Division Method

Area of Science:

• Pharmacology and Cheminformatics
• Computational Drug Discovery
• Biotechnology

Background:

• Drug metabolism by cytochrome P450 (CYP450) isoforms is complex and impacts drug efficacy and safety.
• Preventing drug-drug interactions is a critical challenge in novel pharmaceutical development.
• Accurate prediction of CYP450 enzyme-substrate selectivity is essential for drug design.

Purpose of the Study:

• To develop and evaluate computational models for predicting CYP450 enzyme-substrate selectivity.
• To compare the performance of various feature sets and machine learning models for this prediction task.
• To introduce and assess novel network-based label space division (NLSD) methods in the context of drug metabolism.

Main Methods:

• Formulated CYP450 enzyme-substrate selectivity prediction as a multilabel learning problem.
• Evaluated feature combinations including physiochemical properties, mol2vec, and molecular fingerprints (ECFP, MACCS).
• Applied and compared seven multilabel models: ML-kNN, multilabel twin SVM, and five NLSD-based methods (NLSD-MLP, NLSD-XGB, NLSD-EXT, NLSD-RF, NLSD-SVM).

Main Results:

• The NLSD-based methods, especially NLSD-XGB, demonstrated superior performance compared to previous work.
• NLSD-XGB achieved high prediction success rates: 91.1% (top-1), 96.2% (top-2), and 98.2% (top-3).
• Significant improvements were observed with NLSD-XGB over existing methods, exceeding 11% and 14% in top-1 accuracy via cross-validation and hold-out tests, respectively.

Conclusions:

• Network-based label space division models represent a novel and effective approach for predicting drug metabolism.
• The NLSD-XGB model offers a significant advancement in predicting CYP450 enzyme-substrate selectivity.
• This study provides a robust computational tool to aid in the development of safer pharmaceuticals by minimizing drug-drug interactions.