Discovering TEAD4 modulators for hepatocellular carcinoma: a GAN-enabled generative modelling framework

Introduction Liver diseases continue to impose a major global health burden, and therapeutic progress is constrained by the limited availability of validated small-molecule modulators. TEAD4, a central Hippo-YAP effector, has emerged as a key regulator of hepatic regeneration, survival, and disease progression, yet remains pharmacologically underexplored due to the scarcity of experimentally confirmed inhibitors. Critically, the limited number of known active compounds restricts effective supervised learning, necessitating data augmentation strategies capable of expanding TEAD4 relevant chemical space. Methods To address this, we developed an integrative computational framework in which a conditional generative adversarial network was trained on QikProp-derived molecular descriptors to generate chemically realistic synthetic samples and mitigate class imbalance. This GAN driven augmentation enabled construction of a robust activity prediction model. XGBoost was selected as the classifier due to its strong performance on structured descriptor datasets and its ability to capture complex nonlinear relationships with strong generalization. The augmented dataset was used to train the XGBoost classifier for activity prediction and screen DrugBank compounds, producing a focused set of high confidence candidates. Shortlisted hits were refined using structure-based evaluation, toxicity filtering, and anticancer sensitivity prediction. Results Quantum chemical analysis identified DB00169 (cholecalciferol) as a potential TEAD4-binding candidate supported by combined structural, dynamic, and electronic analyses. Molecular dynamics simulations further supported the stability of the TEAD4–ligand complex, indicating compact structural behaviour and thermodynamically favourable conformational states. Discussion Overall, this work demonstrates that coupling GAN based molecular augmentation with XGBoost classification and molecular simulations provides a scalable strategy for identifying biologically meaningful TEAD4 modulators, supporting TEAD4 targeted drug discovery across liver diseases.