Repeat-dose toxicity prediction with Generalized Read-Across (GenRA) using targeted transcriptomic data: A...

Area of Science:

• Computational toxicology
• In silico methods for chemical safety assessment

Background:

• Read-across is a key method for predicting chemical toxicity using data from similar compounds.
• Generalized Read-Across (GenRA) automates predictions using chemical and bioactivity fingerprints.
• The impact of biological similarity on read-across performance requires further investigation.

Purpose of the Study:

• To assess the influence of biological similarities on neighborhood formation in GenRA.
• To evaluate the performance of GenRA in predicting chemical hazard using transcriptomic data.
• To compare the effectiveness of chemical fingerprints, transcriptomic fingerprints, and hybrid approaches for toxicity prediction.

Main Methods:

• Utilized targeted transcriptomic data (93 genes) for 1060 chemicals in HepaRG™ cells.
• Calculated transcriptomic similarity using binary hit-calls from concentration-response data.
• Evaluated GenRA performance using Area Under the Receiver Operating Characteristic curve (AUC) for US EPA ToxRefDB v2.0 hazard outcomes.

Main Results:

• Modest improvements in ROC AUC scores (2.1% for transcriptomic, 7.3% for hybrid) were observed across all endpoints.
• Significant improvements were noted for liver-specific toxicity endpoints, with ROC AUC scores increasing by 10% (transcriptomic) and 17% (hybrid).

Conclusions:

• Hybrid descriptors combining chemical and targeted transcriptomic data offer improved in vivo toxicity predictions.
• Transcriptomic information enhances the accuracy of automated read-across, particularly for organ-specific toxicities.
• The study highlights the value of integrating multi-modal data for robust chemical safety assessments.