Machine Learning on Toxicogenomic Data Reveals a Strong Association Between the Induction of Drug-Metabolizing Enzymes and Centrilobular Hepatocyte Hypertrophy in Rats

Centrilobular hepatocyte hypertrophy is frequently observed in animal studies for chemical safety assessment. Although its toxicological significance and precise mechanism remain unknown, it is considered an adaptive response resulting from the induction of drug-metabolizing enzymes (DMEs). This study aimed to elucidate the association between centrilobular hepatocyte hypertrophy and DME induction using machine learning on toxicogenomic data. Utilizing publicly available gene expression data and pathological findings from rat livers of 134 compounds, we developed six different types of machine learning models to predict the occurrence of centrilobular hepatocyte hypertrophy based on gene expression data as explanatory variables. Among these, a LightGBM-based model demonstrated the best performance with an accuracy of approximately 0.9. With this model, we assessed each gene's contribution to predicting centrilobular hepatocyte hypertrophy using mean absolute SHAP values. The results revealed that <i>Cyp2b1</i> had an extremely significant contribution, while other DME genes also displayed positive contributions. Additionally, enrichment analysis of the top 100 genes based on mean absolute SHAP values identified "Metabolism of xenobiotics by cytochrome P450" as the most significantly enriched term. In conclusion, the current results suggest that the induction of multiple DMEs, including CYP2B1, is crucial for the development of centrilobular hepatocyte hypertrophy.