Machine learning-based prediction of SARS-CoV-2 bioactivity: integrating IC₅₀ regression and activity classification...

Area of Science:

• Computational chemistry
• Drug discovery
• Machine learning

Background:

• Accurate prediction of compound bioactivity is crucial for accelerating antiviral drug discovery.
• Machine learning (ML) methods show promise in modeling structure-activity relationships and compound potency.

Purpose of the Study:

• To develop an integrated ML framework for predicting IC50 and pIC50 values of SARS-CoV-2 antiviral compounds.
• To support compound prioritization through classification and explore ligand efficiency for activity prediction.

Main Methods:

• Developed a regression model for quantitative IC50 prediction.
• Implemented a classification model for categorizing compounds into active and inactive classes.
• Utilized a multi-task neural network for joint regression and classification, incorporating ligand efficiency.

Main Results:

• The neural network with feature selection achieved a coefficient of determination (R²) of 0.77 for IC50 prediction.
• The Random Forest classifier demonstrated high performance with accuracy, precision, and recall of approximately 0.92.
• The integrated framework showed strong predictive capabilities for SARS-CoV-2 bioactivity.

Conclusions:

• Integrated regression, classification, and multi-task learning approaches offer scalable and cost-effective tools for SARS-CoV-2 bioactivity prediction.
• The proposed ML framework can significantly accelerate antiviral drug discovery and reduce experimental costs.
• Incorporating ligand efficiency provides a novel perspective for compound prioritization in SARS-CoV-2 research.