Classification of HCV NS5B polymerase inhibitors using support vector machine.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Virology

Background:

• Hepatitis C virus (HCV) NS5B polymerase is a key target for antiviral drug development.
• Non-nucleoside analogue inhibitors (NNIs) targeting the NNI III binding site are crucial for HCV treatment.
• Predictive modeling can accelerate the discovery of effective HCV inhibitors.

Purpose of the Study:

• To build and evaluate classification models for predicting the activity of HCV NS5B polymerase inhibitors.
• To identify key molecular descriptors that influence inhibitor binding to the NS5B polymerase.
• To optimize predictive accuracy for active versus weakly active NNIs.

Main Methods:

• Support Vector Machine (SVM) algorithm was employed to construct three classification models.
• Global, 2D, and 3D property autocorrelation descriptors were calculated using ADRIANA.Code for 386 HCV NS5B polymerase NNIs.
• Models were validated using test sets and an external test set.

Main Results:

• Model 2, utilizing 16 global and 2D autocorrelation descriptors, achieved 88.24% prediction accuracy and a Matthews correlation coefficient (MCC) of 0.789 on the test set.
• Model 1, based on 13 global descriptors, demonstrated 86.25% accuracy and 0.732 MCC on an external test set of 80 compounds.
• Key molecular properties influencing ligand-polymerase interactions included shape descriptors, rotatable bonds, and water solubility.

Conclusions:

• Machine learning models, particularly those incorporating global and 2D autocorrelation descriptors, can effectively predict the activity of HCV NS5B polymerase inhibitors.
• Specific molecular descriptors play significant roles in the binding affinity and inhibitory potential of NNIs.
• This approach aids in the rational design and discovery of novel antiviral agents against HCV.