A Neural-Network-Based\nMapping and Optimization Framework\nfor High-Precision Coarse-Grained Simulation

The accuracy and efficiency of a\ncoarse-grained (CG)\nforce field\nare pivotal for high-precision molecular simulations of large systems\nwith complex molecules. We present an automated mapping and optimization\nframework for molecular simulation (AMOFMS), which is designed to\nstreamline and improve the force field optimization process. It features\na neural-network-based mapping function, DSGPM-TP (deep supervised\ngraph partitioning model with type prediction). This model can accurately\nand efficiently convert atomistic structures to CG mappings, reducing\nthe need for manual intervention. By integrating bottom-up and top-down\nmethodologies, AMOFMS allows users to freely combine these approaches\nor use them independently as optimization targets. Moreover, users\ncan select and combine different optimizers to meet their specific\nmission. With its parallel optimizer, AMOFMS significantly accelerates\nthe optimization process, reducing the time required to achieve optimal\nresults. Successful applications of AMOFMS include parameter optimizations\nfor systems such as POPC and PEO, demonstrating its robustness and\neffectiveness. Overall, AMOFMS provides a general and flexible framework\nfor the automated development of high-precision CG force fields.