AI-driven discovery of natural product-derived FAK1 inhibitors for idiopathic pulmonary fibrosis.

Idiopathic pulmonary fibrosis (IPF) is a chronic interstitial lung disease of unknown aetiology with high mortality. Focal adhesion kinase 1 (FAK1) has emerged as a key therapeutic target due to its role in exacerbating pulmonary fibrosis through pathways such as transforming growth factor-β (TGF-β) signalling. Although several anti-fibrotic drugs targeting FAK1 are currently in development, therapeutic outcomes remain suboptimal with numerous limitations. This study established an efficient virtual screening workflow integrating machine learning and deep learning to systematically mine 25,000 natural compounds sourced from TCMBank and HERB. Multiple multi-fingerprint-multi-algorithm combination models were trained using the ChEMBL active compound dataset, identifying the optimal pIC50 prediction model. Key molecular fragments were then characterised using SHAP analysis. Further validation using activity/decoy sets revealed that the PLANET and KarmaDock deep learning docking methods demonstrated favorable enrichment performance for the FAK1 target. Finally, ADMET prediction and molecular dynamics simulations identified six candidate compounds derived from traditional Chinese medicine that stably bind to key residues of FAK1 and exhibit favorable pharmacokinetic properties. Although these results are based on computational predictions and have not yet been validated by in vitro or in vivo experiments, the screening strategy proposed in this study provides an efficient and rapid framework for the large-scale identification of natural products. It offers a theoretical foundation and actionable leads for the future experimental validation of natural FAK1 inhibitors, thereby providing a new avenue for targeted IPF therapy.