Predicting chemical ocular toxicity using a combinatorial QSAR approach.

Area of Science:

• Toxicology
• Computational Chemistry
• In Silico Methods

Background:

• Regulatory agencies mandate chemical and product testing to prevent consumer and worker eye injuries.
• Traditional animal screening methods, like the rabbit Draize test, are time-consuming and expensive.
• Virtual screening using computational models presents an attractive alternative for identifying potential ocular toxicants.

Purpose of the Study:

• To develop and validate quantitative structure-activity relationship (QSAR) models for predicting chemical-induced eye irritation.
• To establish computational tools for screening chemical libraries and prioritizing compounds for further in vivo testing.

Main Methods:

• Utilized a curated dataset of 75 small molecules with existing animal ocular toxicity data.
• Applied k-nearest neighbor and random forest statistical approaches with Dragon and Molecular Operating Environment descriptors.
• Validated the developed QSAR models on an independent external dataset of 34 compounds.

Main Results:

• Individual QSAR models achieved external correct classification rates (CCR) ranging from 72% to 87%.
• A consensus model, averaging predictions from individual models, demonstrated improved performance with a CCR of 0.93.
• The validated models show significant potential for accurate prediction of ocular toxicity.

Conclusions:

• Developed QSAR models provide a reliable in silico method for assessing ocular toxicity.
• These validated computational models can significantly reduce the need for animal testing in chemical safety evaluations.
• The models can be effectively used to screen large chemical libraries and prioritize compounds for targeted in vivo studies.