Chemoinformatics and Machine Learning Approaches for Identifying Antiviral Compounds.

Area of Science:

• Computational chemistry and cheminformatics
• Machine learning applications in drug discovery
• Antiviral research and development

Background:

• Pandemics necessitate rapid development of antiviral drugs and vaccines.
• Computational approaches are crucial for accelerating drug discovery and repurposing.
• Need for predictive models to identify potent antiviral compounds with minimal side effects.

Purpose of the Study:

• To develop and benchmark machine learning models for predicting antiviral compound activity.
• To assess the potential of compounds for managing viral outbreaks.
• To identify effective feature selection methods and machine learning algorithms for antiviral drug discovery.

Main Methods:

• Compiled a dataset of 2358 antiviral compounds from the CAS COVID-19 antiviral SAR dataset.
• Computed 1157 two-dimensional molecular descriptors and selected the most relevant ones using feature selection techniques (Tree-based, Correlation-based, Mutual information-based).
• Benchmarked seven machine learning algorithms: Random Forest, XGBoost, Support Vector Machine, KNN, Decision Tree, MLP Classifier, and Logistic Regression.

Main Results:

• Random Forest and XGBoost models demonstrated superior performance across all feature selection methods.
• Internal validation achieved a maximum predictive accuracy of 88% for both Random Forest and XGBoost models.
• External validation showed a maximum accuracy of 93.10% for XGBoost and 100% for Random Forest.

Conclusions:

• Machine learning, particularly Random Forest and XGBoost, effectively predicts antiviral compound activity.
• High accuracy achieved in both internal and external validation highlights the robustness of the developed models.
• Scaffold analysis provides insights into structure-activity relationships, supporting data-driven antiviral drug discovery.