Bist, P. S.; Bhattarai, S.; Tayara, H.; Chong, K. T.

This work introduces AntiCPs-CompML, a novel Machine learning framework for the rapid identification of anticoronavirus peptides (ACPs). ACPs, acting as viral shields, offer immense potential for COVID-19 therapeutics. However, traditional laboratory methods for ACP discovery are slow and expensive. AntiCPs-CompML addresses this challenge by utilizing three primary features for peptide sequence analysis: Amino Acid Composition (AAC), Pseudo Amino Acid Composition (PAAC), and Composition-Transition-Distribution (CTD). The framework leverages 26 different machine learning algorithms to effectively predict potential anti-coronavirus peptides. This capability allows for the analysis of vast datasets and the identification of peptides with hallmarks of effective ACPs. AntiCPs-CompML boasts unprecedented speed and cost-effectiveness, significantly accelerating the discovery process while enhancing research efficiency by filtering out less promising options. This method holds promise for developing therapeutic drugs for COVID-19 and potentially other viruses. Our model demonstrates strong performance with an F1 Score of 92.12% and a Roc AUC of 76% in the independent test dataset. Despite these promising results, we are continuously working to refine the model and explore its generalizability to unseen datasets. Future enhancements will include feature-based and oversampling augmentation strategies addressing the limitation of anti-covid peptide data for comprehensive study, along with concrete feature selection algorithms, to further refine the predictive power of the models. AntiCPs-CompML ushers in a new era of expedited anti-covid peptides discovery, accelerating the development of novel antiviral therapies. Index Terms: Antiviral peptides, Drug Discovery, Sars-Cov-2, Machine Learning