Identification of a natural product derived NaV1.5 sodium channel modulator through integrated machine learning and molecular modeling

Atrial fibrillation (AF) is the most common cardiac arrhythmia, and the voltage-gated sodium channel NaV1.5 represents a key therapeutic target. However, currently available sodium channel blockers are associated with proarrhythmic liabilities, underscoring the need for safer alternatives. This study aimed to develop an integrated computational framework to identify novel natural product–derived NaV1.5 modulators. A curated dataset of 1512 NaV1.5 inhibitors from ChEMBL was used to train LightGBM, XGBoost, and Random Forest classification models using optimized hybrid feature selection strategies. Model performance was evaluated using scaffold-based splitting and Matthews correlation coefficient (MCC). The optimized models were applied to virtually screen 100,404 natural compounds from the COCONUT database. Top-ranked hits were further analyzed using molecular docking, 100-ns molecular dynamics simulations, MM/GBSA binding free energy calculations, and in silico ADMET profiling. The best-performing model (LightGBM) achieved an MCC of 0.66 on the external scaffold-based test set. Virtual screening identified 10 candidate compounds. Among them, Piperaduncin C exhibited favorable docking scores, stable interactions during molecular dynamics simulations, and a more favorable predicted binding free energy compared to quinidine. ADMET analysis suggested comparatively low predicted hERG liability and limited CYP2D6 inhibition. This integrated computational strategy identified Piperaduncin C as a potential NaV1.5 modulator. While experimental validation is required, these findings support its consideration as a natural product–derived scaffold for further antiarrhythmic drug development.