Designing novel FAK inhibitors targeting gastric cancer: a combined approach using machine learning, docking analysis, molecular dynamics simulations, and experimental validation.

Gastric cancer, the fifth most common cancer worldwide, causes over 650,000 deaths each year. Although targeted therapies have shown effectiveness in advanced stages, their success is often limited by side effects and resistance mechanisms. Focal adhesion kinase (FAK), which is overexpressed in gastric cancer and linked to poor prognosis, has emerged as a promising therapeutic target due to its roles in tumor growth, metastasis, and drug resistance. Despite encouraging preclinical results, FAK inhibitors have not yet gained clinical approval, highlighting the need for new drug discovery methods. In this study, we combined machine learning (ML), molecular docking, and molecular dynamics (MD) to screen for FAK inhibitors systematically. Bioinformatics analysis confirmed FAK overexpression in gastric cancer tissues. A dual ML approach was used: a high-performance classification model (accuracy: 0.9616, precision: 0.9617, F1-score: 0.9613) identified potential FAK inhibitors, while a LightGBM-based regression model (R2 = 0.726, MAE = 0.439, RMSE = 0.632) predicted pIC50 values for the AGS cell line. Virtual screening of 1.6 million compounds resulted in 47,848 candidates with docking scores ≤ -8.00 kcal/mol, of which 10 active inhibitors were selected using ML. Clustering and MD simulations verified stable FAK binding, and in vitro testing identified compound A4 as an active inhibitor with notable anti-tumor activity. This combined computational and experimental approach provides an efficient framework for discovering new FAK inhibitors. It offers a strong basis for future structural optimization, mechanistic research, and in vivo studies in gastric cancer treatment.