Hydrolysis Reaction Rate Prediction Using Machine Learning: WaterDRoP.

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Machine Learning

Background:

• Sustainable chemical design requires predicting the environmental fate of novel compounds.
• Hydrolysis is a critical degradation pathway influencing contaminant behavior in water and biological systems.

Purpose of the Study:

• Develop a machine learning model, WaterDRoP, to predict chemical hydrolysis rates from molecular structure.
• Enable accurate prediction of pollutant degradation potential for compounds lacking experimental data.

Main Methods:

• A two-stage neural network model was trained on 808 experimental hydrolysis rates.
• The model classifies compounds as stable or unstable and estimates half-lives.
• Shapley Additive Explanations (SHAP) were used for atom-level attribution analysis.

Main Results:

• WaterDRoP demonstrates superior performance compared to existing models (EPI Suite, Hydrolysis QSAR, QSAR Toolbox).
• The model achieves high accuracy in stability classification (F1 score) and rate prediction (RMSE, MAE, R²).
• SHAP analysis identified key substructures influencing hydrolysis predictions, aligning with known mechanisms.

Conclusions:

• WaterDRoP provides a reliable in silico tool for estimating hydrolysis rates and predicting contaminant fate.
• The model advances sustainable chemical design by enabling early assessment of degradation potential.
• The tool and curated dataset are openly available to the scientific community.