An Integrated Machine Learning Model and Induced‐Fit Docking for Identifying EGFR Inhibitors from Marine Natural Products

This study aimed to develop machine learning models to predict epidermal growth factor receptor (EGFR) inhibitory activity and apply the best models to screen marine natural products, followed by induced‐fit molecular docking analysis. A total final dataset 13 700 classified as active (IC 50 ≤ 50 n m ) and inactive (IC 50 > 1000 n m ). Molecular fingerprints and RDKit descriptors were used as features, and the dataset was split into 80% training and 20% testing using stratified sampling, with normalization applied. Five algorithms i.e. Random Forest, Logistic Regression, Gradient Boosted Trees, kNN, and Naïve Bayes were evaluated, with Random Forest and kNN achieved highest AUC and statistical evaluation values. Subsequently, these two models were applied in a sequential consensus strategy to screen 31 561 marine compounds from CMNPD database. Compounds predicted as active by all two models were prioritized and subjected to induced‐fit docking in MOE 2014 against EGFR (PDB ID: 1M17). The docking protocol was validated by re‐docking the native ligand (RMSD < 2 Å), and the best hits were analyzed based on binding energy and interaction patterns. CMNPD28093, CMNPD28092, and CMNPD16800 demonstrated strong potential as a candidates compound of EGFR. This integrated approach successfully identified promising marine‐derived candidates for EGFR inhibitor discovery.