Deep convolutional neural networks for generating atomistic configurations of multi-component macromolecules from coarse-grained models

Abstract

Despite the modern advances in the available computational resources, the length and time scales of the physical systems that can be studied in full atomic detail, via molecular simulations, are still limited. To overcome such limitations, coarse-grained (CG) models have been developed to reduce the dimensionality of the physical system under study. However, to study such systems at the atomic level, it is necessary to re-introduce the atomistic details into the CG description. Such an ill-posed mathematical problem is typically treated via numerical algorithms, which need to balance accuracy, efficiency, and general applicability. Here, we introduce an efficient and versatile method for backmapping multi-component CG macromolecules of arbitrary microstructures. By utilizing deep learning algorithms, we train a convolutional neural network to learn structural correlations between polymer configurations at the atomistic and their corresponding CG descriptions, obtained from atomistic simulations. The trained model is then utilized to get predictions of atomistic structures from input CG configurations. As an illustrative example, we apply the convolutional neural network to polybutadiene copolymers of various microstructures, in which each monomer microstructure (i.e., cis-1,4, trans-1,4, and vinyl-1,2) is represented as a different CG particle type. The proposed methodology is transferable over molecular weight and various microstructures. Moreover, starting from a specific single CG configuration with a given microstructure, we show that by modifying its chemistry (i.e., CG particle types), we are able to obtain a set of well equilibrated polymer configurations of different microstructures (chemistry) than the one of the original CG configuration.

Perspective: Advances, Challenges, and Insight for Predictive Coarse-Grained Models

“…Doxastakis and co-workers trained a generative adversarial network (GAN) to perform deterministic back-mapping of polymer configurations in analogy to super-resolution image reconstruction, in which the CG and AA configurations correspond to low- and high-resolution images, respectively . Similarly, Harmandaris and co-workers recently trained a convolutional neural network to predict AA configurations for polymer chains by predicting atomic bond vectors conditioned upon the CG coordinates and chemistry of the corresponding monomer . After relaxing local intermolecular interactions, the resulting polymer melt quite accurately matched the equilibrium structural and thermodynamic properties of the AA polymer model.…”

Integrating Machine Learning in the Coarse-Grained Molecular Simulation of Polymers

“…Li and coworkers 261 implemented a method based on convolutional conditional generative adversarial networks (GANs), following an image representation of red−green−blue values extracted by the systems position vectors and using cis-1,4-polyisoprene melt as test system. In a subsequent work 262 they developed a backmapping strategy that makes use of convolutional autoencoder neural networks. They trained the models to learn to reconstruct the atomistic detail at the level of individual monomers, accounting for their type (each monomer was mapped to a CG bead), using the distribution of bond lengths conditioned on the CG position as target.…”

“…(Upper line) Schematic representation of the strategy implemented by Christofi et al to decode CG systems of polybutadiene random copolymers consisting of three monomer types: cis-1,4, trans-1,4, and vinyl-1,2 monomers. (Lower line) Details of the monomer description used: blue indicate the bond vectors, green indicate the distances between the atomistic particles q i (a united atoms representation was considered) with respect to the center of mass of the monomer Q (in red), q 0 denotes the coordinates of the last united atom particle of the previous monomer along the chain.…”

“…(Lower line) Details of the monomer description used: blue indicate the bond vectors, green indicate the distances between the atomistic particles q i (a united atoms representation was considered) with respect to the center of mass of the monomer Q (in red), q 0 denotes the coordinates of the last united atom particle of the previous monomer along the chain. Reproduced with permission from ref . Copyright 2022, AIP Publishing.…”

Pair-Variational Autoencoders for Linking and Cross-Reconstruction of Characterization Data from Complementary Structural Characterization Techniques

“…An autoencoder is typically trained by minimizing the mismatch (i.e., loss) between the reconstructed data and the input data. Materials scientists have applied autoencoders for multiple purposes such as reconstruction of experimental characterization data, − molecular structure, − or microscopy images; ,,, clustering , and/or classification ,,, of the latent space representations; obtaining material design parameters , or deriving order parameters − from latent space representations; and molecular or material property optimization based on the latent space representations. Some of the studies mentioned here − ,,,, used a modified version of the autoencoder called a variational autoencoder (VAE) that maps the encoded latent space to a multidimensional standard Gaussian distribution, which has the benefit of a continuous latent space compared to the sparse latent space that one would get from the encodings of an unmodified autoencoder.…”