MAPLE: A Machine-Learning Force-Field-Native Platform for Automated Reaction Modeling and Enzyme Design

Abstract

Machine-learning force fields (MLFFs) are reshaping computational chemistry and biology by delivering near-quantum mechanical accuracy at a computational cost comparable to conventional force fields, enabling applications in biomolecular simulation, catalysis, and materials science. However, despite these advances, a unified and automated computational platform enabling the broader application of MLFFs is still lacking. Here, we present MAPLE (MAchine learning Potential for Landscape Exploration), a computational toolkit specially developed for MLFF-based molecular modeling, featuring a tailored software framework and parallelized algorithms for large-scale and versatile molecular modeling tasks. We demonstrated the robustness and usability of MAPLE through systematic benchmarking of state-of-the-art reactive MLFFs and applications to multiple biocatalytic scenarios, highlighting its capability for fast yet accurate simulation of catalytic reactions. By integrating accurate and efficient MLFFs with parallelized algorithms in a highly optimized and flexible software framework, MAPLE serves as a next-generation, physically informed, machine-learning-driven molecular modeling platform with broad applicability to rational catalyst design and drug discovery.