Development of Machine Learning-Based Models for Mutagenicity Predictions with Applications to Non-Sugar Sweeteners.

Area of Science:

• Computational chemistry
• Toxicology
• Food safety science

Background:

• Artificial sweeteners, or non-sugar sweeteners (NSSs), have been used since WWII.
• Concerns exist regarding the mutagenicity potential of NSSs, necessitating safety evaluations for new chemical registrations.
• Traditional in vivo mutagenicity testing is time-consuming and costly.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting the mutagenicity of NSSs.
• To provide a more efficient alternative to experimental mutagenicity testing.
• To identify potentially mutagenic NSSs for further investigation.

Main Methods:

• Development of ML models using a dataset of 6881 organic compounds with random data splitting (50/50).
• Cross-validation analysis to select the best performing ML models.
• Consensus predictions for 332 NSSs using six selected models, incorporating applicability domain assessment.
• Comparison with k-nearest neighbor and toxicity estimation software predictions.

Main Results:

• Six ML models were selected based on stringent cross-validation.
• Consensus predictions identified six compounds as potentially mutagenic NSSs.
• Model-derived predictions showed reliability when compared to other computational methods.

Conclusions:

• ML models offer a viable and efficient approach for predicting NSS mutagenicity.
• The study successfully prioritized six compounds as potentially mutagenic NSSs.
• Developed models are available for public use to aid in food safety assessments.