Accelerating the pace of ecotoxicological assessment using artificial intelligence

Area of Science:

• Environmental Toxicology
• Computational Chemistry
• Ecotoxicology

Background:

• Species Sensitivity Distributions (SSDs) are crucial for assessing chemical ecotoxicological risks.
• Limited experimental toxicity data restricts the development of SSDs for most chemicals.
• Existing SSDs cover only a small fraction of chemicals, hindering comprehensive risk assessment.

Purpose of the Study:

• To develop a novel approach for expanding the chemical coverage of SSDs.
• To leverage Artificial Neural Networks (ANN) to predict ecotoxicity data.
• To generate a comprehensive SSD database for screening-level ecotoxicological assessments.

Main Methods:

• Collected over 2000 experimental Lethal Concentration 50 (LC50) toxicity data points for 8 aquatic species.
• Trained Artificial Neural Network (ANN) models for each species based on molecular structure to predict LC50 values.
• Applied predicted LC50 values to fit SSD curves using a bootstrapping method for 8424 chemicals.

Main Results:

• Developed ANN models with R-squared values ranging from 0.54 to 0.75 (median R² = 0.69).
• Successfully generated SSDs for 8424 chemicals from the ToX21 database.
• Demonstrated a significant expansion of chemical coverage for SSD analysis.

Conclusions:

• The ANN-based approach effectively expands the availability of SSDs for a large number of chemicals.
• The generated SSD database provides a valuable screening-level resource for ecotoxicological impact assessment.
• This method addresses the data scarcity issue, enabling broader chemical risk evaluation.