Drug Discovery Proposal for Influenza B Based on Machine Learning Models

Antimicrobial resistance (AMR) poses a growing global health threat, exacerbated by the improper use of antimicrobial agents. Among viral pathogens, Influenza B stands out due to its role in seasonal epidemics, particularly affecting children, the elderly, and immunosuppressed individuals. Its ability to evade immune responses through antigenic drift and the limited availability of specific inhibitors contribute to its persistence and public health impact. This study presents a machine learning-based approach to predict the inhibitory activity (pIC50) of chemical compounds against the Influenza B virus. Biological activity data (IC50) were obtained from the ChEMBL database and processed into pIC50 values. Molecular descriptors and fingerprints were computed using RDKit from SMILES representations of compounds. Molecular descriptors, fingerprints, and compounds' representations were used to characterize the inhibition activity. This information was used as input for RDKit from SMILES representations of compounds. Several regression models were tested, with the Gradient Boosting Regressor demonstrating the highest performance. The model achieved an $\mathbf{R}^{2}$ of 0.9937, MAE of 0.0369, and strong Pearson and Spearman correlations, indicating high predictive precision and robustness. These findings contribute to computational virology by supporting the faster identification of promising antiviral candidates against Influenza B.