Classification models of HCV NS3 protease inhibitors based on support vector machine (SVM)

Area of Science:

• Medicinal Chemistry
• Computational Biology
• Virology

Background:

• Hepatitis C virus (HCV) infection is a global health concern.
• Inhibiting the HCV non-structural protein 3 (NS3) serine protease is a key therapeutic strategy.
• Developing effective HCV NS3 protease inhibitors requires understanding structure-activity relationships.

Purpose of the Study:

• To build and evaluate machine learning models for predicting HCV NS3 protease inhibitors.
• To identify key molecular descriptors and structural features correlated with inhibitor bioactivity.
• To establish a computational method for virtual screening of novel HCV inhibitors.

Main Methods:

• Utilized a dataset of 413 HCV NS3 protease inhibitors.
• Developed four classification models using the support vector machine (SVM) method.
• Employed Kohonen's self-organizing map for dataset splitting and SVMAttributeEval for descriptor selection.
• Performed Extended Connectivity Fingerprint (ECFP_4) analysis for structural feature identification.

Main Results:

• The best performing SVM model achieved high prediction accuracy (90.76%), sensitivity (92.21%), specificity (88.10%), and MCC (0.799).
• Number of rotatable bonds, charge, and electronegativity were identified as important properties influencing inhibitor bioactivity.
• A unique substructure, cyclopropyl with an acylsulfonamide group, was found in active inhibitors.

Conclusions:

• Machine learning models, particularly SVM, can effectively predict HCV NS3 protease inhibitors.
• Computational methods involving dataset partitioning and feature selection enhance virtual screening efficiency.
• The identified molecular features and substructures provide valuable insights for designing novel HCV therapeutics.