Machine Learning-Accelerated Discovery of Proton-Conducting 2D Materials for Proton Exchange Membranes

Two-dimensional (2D) materials emerge as promising alternatives to conventional polymer-based proton exchange membranes (PEMs) due to their high proton conductivity, mechanical robustness, and surface tunability. Here we present an integrated framework combining ab initio molecular dynamics (AIMD) simulations and machine learning (ML) to accelerate the discovery of proton- and hydrogen-transport properties over 866 nonmetallic 2D materials. Three ML models were trained using AIMD-derived permeation barriers from 488 materials, with Random Forest achieving the highest accuracy and revealing structure-property relationships that govern proton transport. Critical descriptors, including proton-atom distance, pore size, interlayer spacing, and electron affinity, emerged as key predictors of permeation behavior. H⁺/H₂ selectivity through additional AIMD simulations allowed identifying 18 promising candidates, including the experimentally studied graphene and hexagonal boron nitride, thus supporting the robustness of our approach. Experimentally synthesized but barely explored materials, including 2D Si, Ge, TeC, TeCl, GeSe and CSe, emerged as strong candidates for proton conducting membranes. The framework further highlights theoretically stable compounds as unexplored opportunities for PEMs. By integrating atomic-scale simulations with data-driven models, this work provides both fundamental insights into proton permeation mechanisms and practical guidance for designing selective, high-performance nanomaterials for hydrogen energy technologies.