Quantifying and comparing the effects of key chemical descriptors on metal–organic frameworks water stability with CatBoost and SHAP

Metal-organic frameworks (MOFs) were considered suitable candidates for a range of industrial applications, including adsorption, separation, sensing and catalysis, due to their advantages of diverse structures and adjustable functions. One of the criteria for determining the commercial viability of MOFs is their stability in water vapor. Here, we established a novel Categorical Boosting (CatBoost) machine learning approach to model more than 200 datasets of empirical measurements of MOF water stability, and used a comprehensive set of chemical descriptors to represent MOF composition including metal ions, organic ligands, and metal–ligand molar ratios. CatBoost algorithm was significantly superior to other gradient algorithms in accuracy, precision and F1-Score. Also, the CatBoost output was interpreted using the Shapley additive interpretation (SHAP) method. Besides providing guidelines for future experimental screening of stable candidates for MOFs, the interpretable Catboost model can also be used for MOFs screening of other design criteria.