Integrating machine learning, deep learning, and docking to predict aristolochic acid A carcinogenesis

Longzhu Li, Jiacheng Liao, Xintian Chen, Zeqiong Lin, Siqiao Gong, Junmin Huang, Ziqian Bi, Tianyang Wang, Xinliang Chia, Lu Chen, Yongzhi Xu, Huafeng Liu, Junfeng Hao, Jiansong Qi

Objective This study investigates the molecular mechanisms of renal clear cell carcinoma (RCC) induced by Aristolochic acid A (AAA) using machine learning, deep learning, and molecular docking approaches. Methods To identify AAA target genes associated with RCC, differential expression analysis was performed on multiple datasets. Network toxicology, machine learning, deep learning, and molecular docking were used to explore the binding interactions between AAA and target proteins. The top candidate gene was validated using molecular dynamics simulation and in vitro Western blot assays. Results A total of 74 genes were identified as potential targets in AAA-induced RCC. Subsequent machine learning analysis identified seven core genes as key regulators of RCC. Deep learning classification further highlighted five of these seven genes, including PYGL, ADH1B, PTGS1, EDNRA, and AURKA. Additionally, molecular docking simulations revealed strong binding affinities between AAA and these target proteins. Molecular dynamics simulation demonstrated the binding stability of the AAA-PYGL complex, and in vitro studies highlighted PYGL as a potential target of AAA. Elevated expression of PYGL was observed in both 786-O and AAA-induced HK-2 cells. Moreover, treatment with CP-91149 (a PYGL inhibitor) or PYGL knockdown restored the expression of E-cadherin, an epithelial-mesenchymal transition (EMT) marker, in HK-2 cells. Conclusion By combining advanced computational methods with in vitro studies, this work elucidates a key toxicity mechanism of AAA in RCC. Our approach provides a feasible and efficient framework for toxicological studies, offering significant value for toxicologists with limited access to clinical specimens.