TCN-RDP: Predicting Drug-Induced Liver Injury from Time-Series Toxicogenomic Data

Drug-induced liver injury (DILI) is a major obstacle in drug development, often leading to high failure rates in clinical trials. Traditional toxicological assessments are slow and resource-intensive, making early prediction of hepatotoxicity a significant challenge. To address this, we propose TCN-RDP, a novel model that integrates Temporal Convolutional Networks (TCN) and Random Dimension Permutation (RDP). TCN captures temporal dependencies in gene expression data, while RDP enhances the modeling of high-dimensional gene interactions, allowing for better feature representation. Additionally, an XGBoost-based gene selection algorithm improves the model's interpretability by focusing on the most relevant genes. In comparison to conventional methods, TCN-RDP achieved a significant accuracy of 84.05%. This model provides a biologically interpretable framework for early stage hepatotoxicity prediction, offering a more efficient and precise approach to drug safety assessment. Future work aims to extend the model to human-derived data and further refine toxicity stratification, enhancing its potential for regulatory decision-making and early drug screening.