MarineTox Predictor: An Online Library Platform for Enhancing Low-Resourced Saltwater Ecotoxicity Prediction via...

Area of Science:

• Environmental Science
• Computational Toxicology
• Marine Biology

Background:

• Extensive freshwater ecotoxicity data exists, but saltwater data is scarce, hindering marine species toxicity prediction.
• Existing models struggle to predict toxicity endpoints across diverse marine organisms due to data limitations.

Purpose of the Study:

• To develop a novel computational tool, MarineTox predictor, for predicting saltwater toxicity endpoints for marine species.
• To address the challenge of limited saltwater ecotoxicity data by integrating cross-domain knowledge sharing.

Main Methods:

• Integrated multitask deep learning with a cross-domain knowledge-sharing mechanism.
• Developed MarineTox predictor for end-to-end prediction of 31 saltwater toxicity tasks across 26 organisms.
• Utilized freshwater ecotoxicity data to improve prediction for low-resourced saltwater tasks.

Main Results:

• MarineTox predictor improved prediction capacity for 18 low-resourced tasks by up to 140% (R^2 of 0.5-0.93).
• Identified six key substructures with toxicity to specific marine organisms by constructing an interaction network.
• Derived environmental hazard thresholds for ~68,000 chemicals, identifying 902 of concern to the marine ecosystem.

Conclusions:

• MarineTox predictor effectively enhances marine ecotoxicity prediction using shared features from freshwater data.
• The tool aids in understanding structure-dependent toxicity mechanisms and identifying hazardous chemicals.
• An online library of 1.2 million records supports comprehensive marine ecological risk assessment.