About DILImap

Why DILI Remains a Critical Challenge in Drug Development

Drug-Induced Liver Injury (DILI) is a one of the most significant barriers in drug development, responsible for costly late-stage failures and market withdrawals. Why do existing preclinical models struggle to adequately predict DILI? DILI arises from complex, multifactorial mechanisms, including mitochondrial dysfunction, oxidative stress, and reactive metabolite formation, with idiosyncratic cases being especially unpredictable. Current preclinical methods, such as QSAR models and in vitro assays, offer limited sensitivity and and fail to capture the underlying complexity of DILI. Despite advances like 3D assays, these approaches remain reductionist and fail to address the underlying causes of DILI. This underscores the urgent need for more comprehensive and predictive models to mitigate late-stage failures and improve drug safety.

Introducing DILImap: Building a Foundation for Toxicogenomics

Toxicogenomics offers a transformative approach to understanding and predicting DILI. By leveraging gene expression signatures indicative of DILI, this method can predict DILI risk and provide insights into the molecular mechanisms of DILI. We believe this method is a potential game-changer as it provides a distinct edge in its predictive performance, comprehensive mechanistic coverage and scalability. DILImap is a comprehensive RNA-seq library featuring 300 compounds tested across various concentrations, built to capture the complexity of DILI mechanisms. It leverages resources like DILIrank and LiverTox to categorize compounds based on their liver injury history. This dataset provides the foundation for uncovering molecular pathways and early gene signatures indicative of liver injury.

Introducing ToxPredictor: A Machine Learning Model for DILI Prediction

ToxPredictor, a machine learning model trained on DILImap, achieves 88% sensitivity in detecting DILI-positive compounds and 100% specificity in identifying safe compounds in blind validation. This performance surpasses 20+ preclinical methods. ToxPredictor uses a random forest algorithm to estimate dose-specific DILI risk probabilities. The model is trained on pathway-level signatures derived from differentially expressed genes identified relative to DMSO controls. Enrichment analysis, performed using WikiPathways, pinpoints biological pathways disrupted by DILI-associated compounds, providing a deeper understanding of the molecular mechanisms at play. By modeling probabilities across different concentrations, ToxPredictor estimates a transcriptomic first DILI dose, defined as the first dose where the predicted probability exceeds 0.7. Safety margins are quantified as the ratio of the transcriptomic first DILI dose to the maximum plasma concentration (Cmax) at therapeutic levels. A safety margin threshold of 80 was established as the optimal cutoff to classify compounds as high/low risk.