Binding Activity Classification of Anti-SARS-CoV-2 Molecules using Deep Learning Across Multiple Assays

Area of Science:

• Computational Biology
• Drug Discovery
• Machine Learning

Background:

• The COVID-19 pandemic necessitates rapid identification of therapeutic solutions.
• Traditional drug discovery methods are time-consuming and labor-intensive.
• Deep learning offers efficient data processing for complex biological insights.

Purpose of the Study:

• To apply deep neural networks (DNN) with SMOTE for enhanced classification of anti-SARS-CoV-2 molecule binding activities.
• To improve the identification of potential drug candidates against SARS-CoV-2.

Main Methods:

• Utilized 11 diverse SARS-CoV-2 bioassay datasets.
• Employed SMOTE to address class imbalance in datasets.
• Developed and optimized a DNN with specific activation functions, batch normalization, and Adam optimization.
• Evaluated model performance using metrics like BACC, precision, recall, F1 score, MCC, and AUC.

Main Results:

• DNN performance varied with compound ratios; robust results were seen in balanced assays (AlphaLISA, CoV-PPE).
• Highly imbalanced assays (3CL, cytopathic effect) showed higher recall but lower precision.
• The DNN model generally performed well, achieving favorable BACC, MCC, and AUC, especially considering dataset imbalance.

Conclusions:

• Deep learning, particularly DNN with SMOTE, significantly enhances COVID-19 drug discovery by improving active compound identification.
• This computational approach effectively handles high-throughput screening data imbalances.
• The study demonstrates the superiority of advanced computational techniques over traditional models in this domain.