Toxicogenomics strategies for predicting drug toxicity

Area of Science:

• Toxicogenomics
• Drug Development
• Computational Biology

Background:

• Drug candidate failure due to toxicity, particularly hepatotoxicity, remains a significant challenge in pharmaceutical development.
• Early assessment of compound toxicity is crucial for efficient drug development pipelines.

Purpose of the Study:

• To explore novel toxicogenomics approaches for understanding and predicting hepatotoxic potential of pharmaceutical compounds.
• To develop and assess an early-stage toxicity screening method for drug candidates.

Main Methods:

• Utilized two commercial knowledge bases merging human drug toxicity data with rat gene expression (hepatocytes and whole rats).
• Employed a stochastic gradient boosting machine learner to build toxicity classification rules.
• Applied clustering methods to analyze compound and gene sets.

Main Results:

• Developed robust classification rules for both in vitro (hepatocytes) and in vivo (liver) toxicity using high-dose, 24-hour data.
• Observed limited overlap in gene lists between in vitro and in vivo classifiers.
• Found that experimental design, including dose and time points, significantly impacts classifier performance.

Conclusions:

• A compound screening system based on developed toxicity classifiers is feasible.
• Classifier operating characteristics can be tuned for specific screening implementations.
• Addressing issues like microarray site variability and generalizability is essential for robust performance.