Natural Product‐Derived ErbB1 Inhibitors Identified Through Machine Learning‐Based QSAR, Molecular Docking, and Molecular Dynamics Simulations

Abstract

Cancer represents a major global health burden, with abnormal ErbB1 (EGFR) signaling implicated in multiple tumors. Despite the clinical availability of several ErbB1 inhibitors, their long‐term efficacy is often limited by drug resistance, adverse effects, and restricted chemical diversity, underscoring the need for novel inhibitory scaffolds. Natural products represent a largely underutilized source of structurally diverse bioactive compounds; however, their systematic exploration against ErbB1 has been hindered by the lack of robust, large‐scale predictive approaches. This work developed an integrated computational strategy to identify novel natural product‐derived ErbB1 inhibitors. A machine learning QSAR classification model based on XGBoost was trained on a curated dataset of 6953 ErbB1 inhibitors, achieving strong predictive performance (accuracy = 85.3%, AUC = 0.92). Applying this model to over 80,000 natural products from the ZINC database yielded a focused subset of high‐confidence candidates. Subsequent molecular docking analyses revealed that five compounds engage with key catalytic residues of ErbB1, notably MET793 and ASP855, in a manner comparable to clinically used inhibitors. Molecular dynamics simulations (100 ns) confirmed stable binding for four candidates, while in silico ADMET evaluations supported favorable drug‐like properties for most hits. Overall, this study identifies promising natural scaffolds for ErbB1 inhibition and provides a scalable computational framework to support future experimental validation and lead optimization.