Molecular docking–QSAR–Kronecker-regularized least squares-based multiple machine learning for assessment and prediction of PFAS–protein binding interactions

Abstract

Ubiquitous per- and poly-fluoroalkyl substances (PFAS) threaten human's health and attract worldwide attention. PFAS-mediated toxicity involves adverse effects of PFAS on proteins, and assessment of PFAS–protein binding interactions helps to explain PFAS’ adverse effects on human health. In-silico modeling can generate information and decrease experimental costs. Accordingly, in this study, molecular docking was used to determine the binding affinities of 430 PFAS with human serum albumin (HSA), peroxisome proliferator-activated receptor gamma (PPARγ), and transthyretin (TTR). Specifically, analytic hierarchy process, fuzzy comprehensive evaluation, and quantitative structure–activity relationship model were used to assess and predict the binding affinities between PFAS and HSA, PPARγ, and TTR. The binding patterns were determined by defining “PEOE_RPC-, E_vdw, MNDO_LUMO, and vsurf features” as key factors related to charge, energy and shape characteristic of PFAS. Finally, Kronecker-regularized least squares (Kron-RLS) model was applied to predict the binding affinities between PFAS– and G protein-coupled receptor 40 (GPR40), as a new target for prediction. Results showed that the Kron-RLS model exhibited good performance and generated precise predictions (R² = 0.94). In conclusion, this study demonstrated that computational simulations could be used to aid the scientific management of the growing number of PFAS, and could be broadened to include a wide range of environmental contaminations.