Machine learning classification of steatogenic compounds using toxicogenomics profiles

Area of Science:

• Toxicology
• Computational Biology
• Genomics

Background:

• New approach methodologies are driving the development of computational models for toxicity testing.
• Transcriptomic data is increasingly utilized to predict chemical-induced adverse effects.
• Drug-induced hepatic steatosis is a significant concern in chemical safety assessment.

Purpose of the Study:

• To apply supervised machine learning to transcriptomic data for predicting drug-induced hepatic steatosis.
• To evaluate the performance of different machine learning classifiers in this prediction task.
• To gain mechanistic insights into the biological processes underlying drug-induced hepatic steatosis.

Main Methods:

• Utilized supervised machine learning on gene expression data from primary human hepatocytes and rat liver models (in vitro and in vivo).
• Evaluated five machine learning classifiers using microarray data from the Open TG-GATEs database.
• Performed functional profiling and enrichment analyses on top-ranked predictive genes.

Main Results:

• Support vector machine (SVM) achieved the highest predictive performance across all models (ROC-AUCs: 0.820 human, 0.975 rat in vitro, 0.966 rat in vivo).
• Enrichment analyses revealed strong associations of predictive genes with lipid metabolism, mitochondrial function, insulin signaling, and oxidative stress.
• Key genes like CYP1A1, PLIN2, and GCK were linked to lipid metabolism and liver disease, while others indicated novel transcriptomic signals.

Conclusions:

• Machine learning models, particularly SVM, effectively predict drug-induced hepatic steatosis using transcriptomic data.
• These models capture biologically relevant signals and offer mechanistic insights into steatosis pathogenesis.
• The SVM model shows promise as a scalable and interpretable tool for chemical risk assessment and advancing non-animal testing approaches.