A Generalized Machine-Learning Framework for Developing Alchemical Many-Body Interaction Models for Polymer-Grafted Nanoparticles

Polymer-grafted nanoparticles (PGNs) serve as highly customizable building blocks for technologically relevant self-assembled nanomaterials. Physics-informed inverse design strategies are crucial for expediting exploration of the massive associated design space by identifying optimal PGN attributes, such as polymer length and grafting density, to meet a target self-assembled structure. However, their success hinges on an accurate description of how PGN interactions vary as a function of both particle positions and physical attributes. We introduce a framework to meet this need. Specifically, we develop an "alchemical" machine-learned interatomic model (ML-IAM) that describes how PGN interactions vary as a function of both inter-PGN distances <i>and</i> tunable PGN attributes, simultaneously. This model is an extension of the physics-informed and explicitly many-bodied ChIMES ML-IAM. The resulting extended ChIMES (X-ChIMES) ML-IAM is trained on potential of mean force (PMF) data for PGNs with varied polymer ligand lengths. We enable efficient training data generation by combining the forward-reverse steered molecular dynamics enhanced sampling approach with a grid-sampling scheme in the HOOMD-blue software package. We demonstrate the efficacy of ChIMES for generating coarse-grained (CG) models for PGNs with fixed design attributes, development of X-ChIMES, and its application for enhancing digital alchemy inverse-design simulations using HOOMD-blue and the ChIMES Calculator. This is the first application of ChIMES in modeling CG systems and coupling with an inverse design method to target nanomaterial self-assembly.