Ensemble Quantitative Read-Across Structure-Activity Relationship Algorithm for Predicting Skin Cytotoxicity.

Area of Science:

• Toxicology
• Computational Chemistry
• cheminformatics

Background:

• Read-across (RA) and quantitative structure-activity relationship (QSAR) are crucial for filling data gaps in chemical safety assessments.
• These methods utilize chemical structure and properties to predict unknown substance behaviors, minimizing animal testing and costs.
• Developing robust predictive models is essential for efficient chemical registration and risk assessment.

Purpose of the Study:

• To develop and validate a stacked ensemble quantitative read-across structure-activity relationship (enQRASAR) algorithm.
• To predict skin irritation toxicity, specifically the negative log cell viability inhibition concentration at 50% (pIC50) against skin keratinocytes.
• To provide a reliable computational tool for assessing chemical cytotoxicity.

Main Methods:

• Development of a stacked ensemble algorithm integrating RA and QSAR principles (enQRASAR).
• Prediction of skin irritation toxicity using pIC50 as the endpoint.
• Validation of the model's goodness-of-fit and predictability using leave-one-out cross-validation and external test datasets.

Main Results:

• The enQRASAR algorithm demonstrated statistically reliable goodness-of-fit, robustness, and predictability.
• The model achieved low prediction error, even when applied to FDA-approved drugs.
• The developed algorithm effectively predicts the skin cytotoxicity of chemicals.

Conclusions:

• The enQRASAR algorithm is a validated and reliable tool for predicting skin cytotoxicity.
• This computational approach significantly aids in reducing the need for animal testing in chemical safety evaluations.
• The publicly available enQRASAR model facilitates toxicity predictions for unknown compounds in chemical registrations.