A k-NN algorithm for predicting the oral sub-chronic toxicity in the rat

Area of Science:

• Toxicology
• Computational Chemistry
• Regulatory Science

Background:

• Repeated dose toxicity studies are crucial for chemical safety assessment, particularly for determining the Lowest-Observed-(Adverse)-Effect-Level (LO(A)EL).
• Current regulatory frameworks like REACH and cosmetic product regulations mandate LO(A)EL evaluation.
• In vivo animal testing has been the standard method for assessing repeated dose toxicity.

Purpose of the Study:

• To develop and validate a computational model for predicting sub-chronic oral toxicity in rats.
• To provide a reliable alternative to in vivo testing for repeated dose toxicity assessment.
• To support regulatory decision-making and the reduction of animal testing.

Main Methods:

• A k-Nearest Neighbors (k-NN) approach was customized for toxicity prediction.
• A training dataset of 254 chemicals was used to build predictive models.
• Model robustness was assessed using leave-one-out cross-validation and external validation on 179 chemicals.

Main Results:

• The developed k-NN models achieved promising predictive performance with q²≥0.632 and external r²≥0.543.
• Restrictive, user-adjustable rules were implemented to enhance prediction confidence and exclude uncertain chemicals.
• The models demonstrated effectiveness despite the heterogeneous nature of the toxicological data.

Conclusions:

• The developed computational approach offers a valuable tool for predicting chemical toxicity and prioritizing safety assessments.
• These findings can aid in making informed decisions regarding chemical safety and justify waiving animal tests.
• The study contributes to advancing non-animal testing strategies in chemical safety evaluations.