A SAR and QSAR study on 3CLpro inhibitors of SARS-CoV-2 using machine learning methods

Area of Science:

• Computational chemistry and cheminformatics
• Drug discovery and medicinal chemistry
• Machine learning in drug design

Background:

• The 3C-like Proteinase (3CLpro) is essential for novel coronavirus replication and a key target for antiviral drug development.
• Understanding structure-activity relationships (SAR) is crucial for designing effective inhibitors.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting 3CLpro inhibitory activity.
• To identify key structural features associated with potent antiviral compounds.
• To analyze the applicability domains of the developed predictive models.

Main Methods:

• Characterization of 889 compounds using ECFP_4 and MACCS fingerprint descriptors.
• Construction and evaluation of 24 classification models using SVM, RF, XGBoost, and DNN algorithms.
• Analysis of model applicability domains using d_STD-PRO calculations.
• K-means clustering of compounds and SAR analysis of active subsets.
• Development of 27 Quantitative Structure-Activity Relationship (QSAR) models for 464 3CLpro inhibitors.

Main Results:

• The DNN- and ECFP_4-based Model 1D_2 achieved high performance with MCC values of 0.796 (cross-validation) and 0.722 (test set).
• QSAR models demonstrated a minimum RMSE of 0.509 on the test set.
• SAR analysis revealed relationships between structural fragments and inhibitory activities.

Conclusions:

• Machine learning, particularly DNNs with ECFP_4 descriptors, shows significant promise for identifying novel 3CLpro inhibitors.
• The developed models and SAR insights can guide the design of new antiviral agents targeting coronaviruses.
• Quantitative Structure-Activity Relationship (QSAR) modeling provides a robust framework for predicting compound activity.