Predictive toxicogenomics in preclinical discovery.

Area of Science:

• Pharmacology
• Toxicology
• Genomics

Background:

• Drug candidate failure in clinical trials due to toxicity, particularly hepatotoxicity, remains a significant challenge in pharmaceutical development.
• Early assessment of drug-induced liver injury is crucial for efficient drug development.

Purpose of the Study:

• To explore predictive toxicogenomics approaches for understanding and assessing hepatotoxic potential in human drug candidates.
• To develop robust classification rules for early toxicity prediction using integrated data.

Main Methods:

• Utilized two commercial knowledgebases with hybrid experimental design: human drug-toxicity data from literature and rat-based gene expression measures (hepatocytes and liver).
• Employed a stochastic gradient boosting machine learner for toxicity classification, with error estimation via modified bootstrap.
• Applied clustering methods based on compounds and genes to analyze relationships.

Main Results:

• Developed robust classification rules for both in vitro (hepatocytes) and in vivo (liver) data using high-dose, 24-h experimental designs.
• Observed minimal overlap in gene lists between hepatocyte and liver classifiers.
• Found that earlier time points and low-dose data did not yield robust classifiers, highlighting the importance of experimental design.

Conclusions:

• Predictive toxicogenomics approaches show feasibility for developing compound screening assays to identify hepatotoxicity.
• Classifier operating characteristics can be tuned for specific preclinical testing implementations.
• Optimized experimental design is critical for successful toxicity prediction models.