HydroFate - A machine learning-based classification modeling platform for the prediction of hydrolytic stability of...

Hydrolytic degradation is a crucial process influencing the environmental persistence and fate of chemical substances. In this research, the authors have employed a machine learning (ML) approach to develop a classification-based quantitative structure-property relationship (QSPR) model using a curated dataset of experimentally reported hydrolysis half-lives and degradation rates collected from the eChemPortal database (https://www.echemportal.org/echemportal/) to predict the hydrolytic degradation potential of organic chemicals across different pH conditions (pH 4, 7, and 9). Molecular descriptors and fingerprints capturing structural, electronic, and physicochemical features were generated using the RDKit module. After the data set division, feature selection was performed using Gini importance (GI), SHapley Additive exPlanations (SHAP) analysis, and most discriminant feature (MDF) selection algorithms. The descriptors common across different feature selection approaches, along with indicator variables denoting the pH environments, were selected to develop models using 10 ML algorithms. Among these, the categorical boosting (CatBoost) model was the best-performing, achieving predictive accuracies of 0.99 and 0.82, and balanced accuracies of 0.99 and 0.82, on the training and test sets, respectively. Model interpretation via SHAP analysis provided mechanistic insights, highlighting the importance of specific molecular substructures and electronic descriptors in governing hydrolytic reactivity. Additionally, a Python-based predictive platform, "HydroFate," has been developed to make the model easily accessible for predicting the hydrolytic potential of organic chemicals. Thus, the developed model can serve as a reliable computational tool for preliminary screening of chemical stability under OECD TG 111 conditions, reducing the need for extensive experimental testing and supporting regulatory decision-making in chemical safety assessment.