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

Area of Science:

• Biomedical Informatics
• Computational Toxicology
• Pharmacogenomics

Background:

• Drug-induced liver injury (DILI) poses a significant challenge in drug development, leading to high clinical trial failure rates.
• Current toxicological assessments are time-consuming and costly, hindering early hepatotoxicity prediction.
• Developing efficient and accurate methods for early DILI detection is crucial for drug safety.

Purpose of the Study:

• To introduce TCN-RDP, a novel computational model for early-stage hepatotoxicity prediction.
• To leverage Temporal Convolutional Networks (TCN) and Random Dimension Permutation (RDP) for improved analysis of gene expression data.
• To enhance the interpretability and accuracy of DILI prediction models.

Main Methods:

• Integration of Temporal Convolutional Networks (TCN) to analyze temporal dependencies in gene expression data.
• Application of Random Dimension Permutation (RDP) to model high-dimensional gene interactions effectively.
• Utilizing an XGBoost-based algorithm for feature selection to identify key genes influencing hepatotoxicity.

Main Results:

• The TCN-RDP model achieved a high prediction accuracy of 84.05% for drug-induced liver injury.
• The model demonstrated improved feature representation and captured complex gene interactions.
• The integrated gene selection algorithm enhanced the biological interpretability of the model.

Conclusions:

• TCN-RDP offers a biologically interpretable and efficient framework for early hepatotoxicity prediction.
• The model presents a more precise approach to drug safety assessment compared to traditional methods.
• Future research will focus on validating the model with human-derived data and refining toxicity stratification for regulatory applications.