Predicting Polymer Grafted Nanoparticles Assembly using Deep Learning

Grafting polymer chains on nanoparticles (NPs)' surfaces is a well-known route to control their self-assembly and distribution in a polymer matrix. Here we propose a DL method to solve this problem. In this method, two deep neural networks (DNNs) learn the effective interaction between a pair of GNPs from the MD trajectory of a cluster of polymer GNPs. The two DNNs represent potential energy surfaces of the centroid and the point of grafting on an NP's surface. We consider a coarse-grained phenomenological model of NPs and polymers to establish and validate the model. CGMD simulation of a cluster of polymer GNP is conducted to generate the training data. We show that the deep learning model predict the energy of polymer GNPs clusters very accurately. We further carry out the deep learning potential of mean force (DL-PMF) based molecular simulation that predicts the self-assembly of many polymer GNPs into various anisotropic superstructures, including percolating networks and bilayers depending on NPsconcentration in 3D. The DL self-assembled structures are consistent with the first principle based actual superstructures of the nanoscopic building blocks. We find this deep learning approach very efficient and accurate in capturing anisotropic interactions, and it predicts long-range anisotropic aggregates of polymer GNPs. We expect that the potential of mean force for systems with atomistic details can be modeled using this DL framework.