Integrating Machine Learning and Molecular Modeling for Prediction of Anti‐Chikungunya Flavonoids

Chikungunya virus (CHIKV) affects millions during epidemics and has no approved antivirals. An emerging method to predict antiviral drugs is to use machine learning (ML) models that use quantitative structure‐activity relationships. This study evaluated different ML models for predicting the anti‐CHIKV activity using a dataset of 613 compounds. Of the nine ML models that were implemented with 10‐fold cross‐validation, extra trees (ET), logistic regression (LR), random forest (RF), gradient boosting (GB), and artificial neural network (ANN) models, showing better performance in sensitivity/specificity, were used for further predictions. The ensemble comprising these models achieved a prediction accuracy of 75% when evaluated on 12 novel compounds with known anti‐CHIKV activity apart from the 10‐fold cross‐validation dataset. Fifteen potential anti‐CHIKV flavonoids of 1171 were shortlisted using an ensemble of the selected ML models, ADMET properties, and virtual screening against the envelope, capsid, and the non‐structural proteins. Molecular dynamics simulations revealed that kievitone, isowighteone, citrusinol, methylophiopogonone A, 6‐prenylapigenin, lupiwighteone and gancaonin G had strong binding affinity and stable interactions with multiple CHIKV targets. The study demonstrates the utility of established computational methods in identifying flavonoids that can be evaluated using in vitro and in vivo experiments, leading to the development of promising therapeutics against CHIKV infections.