ALE interpretability analysis based on machine learning MOFs adsorption of tetracycline antibiotics in water

With the large-scale use of tetracycline antibiotics (TCs), residues of TCs have been detected in various environmental systems. The adsorption of TCs in water by metal-organic framework materials has been proved to be an efficient removal method. In this study, machine learning (ML) method was used to train MOFs to adsorb TCs, and a ML model was obtained to accurately predict the adsorption of TCs by MOFs. Data sets for model training and evaluation were constructed by collecting data from literature. Six regression models, RF, Adaboost, GBDT, XGB, SVR and KNN, were trained to evaluate the model performance. According to the comparison of evaluation parameters, the model with high prediction performance for MOFs adsorption of TCs is XGB model. Accumulated Local Effects Plot (ALE) was used to visualize XGB model predictions. The results show that Zn and Zr as central metals have higher MOFs potential, while H4TBAPy and H4TCPP have higher MOFs advantage as organic ligands. When PS>9 nm, 200 m2 g-1＜SA <1000 m2 g-1, 0.25 cm3 g-1<PV<0.5 cm3 g-1 or PV>1.5 cm3 g-1, the adsorption capacity of MOFs for TCs is higher, and PS should satisfy the following conditions first. This study accelerates the application of MOFs to TCs adsorption, provides a new way to screen and synthesize MOFs with high adsorption capacity for TCs.