A priority control list for LCMs in freshwater food chain by deep learning

Area of Science:

• Environmental Chemistry
• Ecotoxicology
• Computational Chemistry

Background:

• Liquid crystal monomers (LCMs) are globally distributed, posing environmental and human health risks due to their persistence, bioaccumulation, and toxicity (PBT).
• Assessing the PBT effects of numerous LCMs across food chains is crucial for environmental risk management.

Purpose of the Study:

• To develop a comprehensive priority control list for commercial LCMs based on their PBT effects in a freshwater food chain.
• To utilize machine learning and deep learning models for optimizing this list and predicting LCM risks.

Main Methods:

• Employed molecular docking to assess PBT effects for 1431 LCMs across three trophic levels (Daphnia pulex, Danio rerio, Pelecanus crispus).
• Developed and optimized a priority control list using machine learning, specifically a Residual Neural Network (ResNet) deep learning model.
• Utilized SHapley Additive exPlanations (SHAP) for analyzing factors influencing PBT effects.

Main Results:

• A matrix of 1431 LCMs × 3 trophic levels × 3 PBT effects was generated.
• The ResNet model achieved high accuracy (0.84 test, 0.85 validation sets).
• 509 LCMs were identified as high-risk, and lower electronegativity functional groups were found to potentially reduce PBT effects.

Conclusions:

• This study presents the first priority control list for PBT effects of commercial LCMs in freshwater food chains.
• The developed ResNet model accurately predicts PBT risks, aiding environmental monitoring and regulation.
• Findings suggest that chemical structure modifications, like using lower electronegativity groups, can mitigate LCM environmental risks.