Potential for Machine Learning to Address Data Gaps in Human Toxicity and Ecotoxicity Characterization.

Area of Science:

• Environmental Chemistry
• Computational Toxicology
• Data Science

Background:

• Machine learning (ML) is increasingly used to address data gaps in chemical assessments.
• The systematic application and broad potential of ML for chemical data gaps remain underexplored.

Purpose of the Study:

• To systematically evaluate and prioritize chemical parameters for ML-based toxicity characterization.
• To assess the potential of ML to predict data across a wide range of chemicals.

Main Methods:

• Prioritized 38 chemical parameters based on toxicity relevance and data availability for ML prediction.
• Conducted chemical space analysis to evaluate ML prediction potential for diverse chemical structures.
• Assessed the percentage of chemicals with predicted data based on the availability of measured data.

Main Results:

• Identified and prioritized 13 key parameters for ML development, flagging nine with critical data gaps.
• ML approaches show potential to predict data for 8-46% of marketed chemicals.
• Prediction potential is achievable even with only 1-10% of chemicals having measured data.

Conclusions:

• This work provides a systematic framework for prioritizing parameters in ML model development for chemical toxicity.
• Results highlight the significant potential of ML to fill data gaps and broaden chemical toxicity characterization.