Predicting drug-induced hepatotoxicity using QSAR and toxicogenomics approaches

Area of Science:

• Toxicology
• Pharmacology
• Computational Chemistry

Background:

• Quantitative structure-activity relationship (QSAR) modeling and toxicogenomics are key predictive tools in toxicology.
• These methods are typically employed independently, limiting comprehensive analysis.
• Drug-induced liver injury is a significant concern in pharmaceutical development.

Purpose of the Study:

• To evaluate statistical models for predicting drug hepatotoxicity in rats.
• To compare the predictive power of chemical descriptors, toxicogenomics profiles, and hybrid approaches.
• To identify key molecular mechanisms underlying drug-induced liver injury.

Main Methods:

• Utilized a dataset of 127 drugs from the Toxicogenomics Project rat liver microarray database.
• Developed QSAR classification models using chemical descriptors and various statistical methods.
• Built predictive models using toxicogenomics data (gene expression profiles) and hybrid models combining both data types.
• Validated models using 5-fold external cross-validation.

Main Results:

• QSAR models with chemical descriptors alone achieved a correct classification rate (CCR) of 61%.
• Models using only toxicogenomics data (85 selected descriptors) achieved a CCR of 76%.
• Hybrid models combining chemical and toxicogenomics data showed CCRs between 68% and 77%.

Conclusions:

• Toxicogenomics data alone provided higher predictive accuracy for hepatotoxicity than chemical descriptors alone.
• Hybrid models enriched the mechanistic interpretation of drug-induced liver injury.
• Concurrent analysis of chemical features and gene expression changes improves understanding and prediction of liver toxicity.