Prediction of organ toxicity endpoints by QSAR modeling based on precise chemical-histopathology annotations

Area of Science:

• Toxicology
• Computational Chemistry
• Drug Discovery

Background:

• Accurate prediction of drug candidate toxicity from chemical structure is crucial for developing safer medicines.
• Existing methods for toxicity prediction are limited, necessitating novel approaches.
• The MetaTox consortium guided the development of a comprehensive toxicopathology ontology.

Purpose of the Study:

• To create a comprehensive ontology of toxicopathology for 19 organs.
• To populate the ontology with chemical compounds causing specific histopathologies.
• To build quantitative structure-activity relationship (QSAR) models for predicting liver and kidney toxicity.

Main Methods:

• Developed a comprehensive ontology of toxicopathology for 19 organs, classifying terms by type and organ substructure.
• Manually annotated full-text research articles to link chemical compounds with specific histopathologies.
• Generated QSAR models using annotated compound-toxicity associations from rat and mouse experiments to predict liver and kidney toxicity subcategories.

Main Results:

• The QSAR models demonstrated good performance in initial validation, with sensitivity ranging from 0.80-0.96 and specificity from 0.85-1.00.
• Subsequent validation on newly added compounds showed 75-87% sensitivity and 60-78% specificity.
• Models predicting specific liver and kidney toxicities (e.g., necrosis, weight gain, lipid accumulation, nephron injury) performed better than general hepatotoxicity and nephrotoxicity models.

Conclusions:

• The developed ontology and QSAR models provide a valuable tool for predicting drug-induced liver and kidney toxicity from chemical structures.
• These models can significantly aid in guiding experimental drug discovery towards safer pharmaceutical candidates.
• Further development for more complex toxicity endpoints is warranted.