AI-based prediction of drug-gene interactions modulating tight junction integrity: A deep learning framework highlighting multiple therapeutic targets.

Introduction: Tight junctions regulate epithelial and endothelial barrier function, and their dysfunction is linked to diseases such as inflammatory bowel disease, asthma, and cancer. Identifying drug-gene interactions influencing tight junctions is critical for therapeutic development. This study proposes a deep learning-based neural network framework to predict drug-induced modulation of tight junction integrity using multi-omics data. Materials and methods: Transcriptomic data from NCBI GEO underwent preprocessing, with DEGs identified and key hub genes extracted via network analysis. A feedforward neural network was trained using these features, with performance evaluated through AUC, CA, F1-score, precision, recall, and specificity, ensuring robust predictive accuracy. Results: The neural network model achieved an AUC of 0.947, CA of 0.980, and F1-score of 0.969, indicating excellent classification performance. Among the predicted candidates, Cimifugin was highlighted for its modulatory effects on CLDN1; additional candidates included Baicalein and Berberine. Discussion: The deep learning model demonstrated superior predictive power compared to traditional methods, with strong precision and recall metrics. The framework provides a scalable, data-driven solution for predicting drug-induced changes in tight junction function, with significant implications for drug discovery and personalized medicine. Conclusion: This study presents a powerful AI-based approach for discovering drug candidates targeting tight junctions, offering potential therapeutic strategies for diseases involving tight junction disruption.