Toxicity prediction using target, interactome, and pathway profiles as descriptors.

Area of Science:

• Computational toxicology
• Systems biology
• Data science in chemical safety

Background:

• In silico methods are crucial for chemical safety evaluation.
• Data science approaches, including knowledge-based methods, are increasingly important for computational risk assessment.
• Predicting complex toxicity endpoints like hepatotoxicity benefits from a systemic, biological network perspective.

Purpose of the Study:

• To develop and validate a data science-based modeling pipeline for predicting chemical hepatotoxicity.
• To utilize compound connections to biological targets, interactomes, and pathways as predictive descriptors.
• To identify key biological targets and pathways involved in hepatotoxicity.

Main Methods:

• A data science pipeline was developed using compound-target, compound-interactor, and compound-pathway profiles.
• Tree-based models, including random forest, were trained to predict hepatotoxicity.
• Model performance was optimized, and descriptor importance was analyzed for biological relevance.

Main Results:

• The optimized model combining all biological network descriptors achieved an accuracy of 0.766.
• Descriptors related to cytochromes P450, heme degradation, and biological oxidation were highly weighted.
• The involvement of RHO GTPase effectors in hepatotoxicity was identified as a significant factor.

Conclusions:

• The developed systems biology-based pipeline is a valuable tool for toxicity prediction.
• The approach provides novel insights into the mechanisms underlying chemical-induced hepatotoxicity.
• Integrating systems biology data offers predictive performance comparable to or exceeding traditional chemical structure-based methods.