Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Thermodynamic Transferability in Coarse-Grained Force Fields using Graph Neural Networks

Emily Shinkle, Aleksandra Pachalieva, Riti Bahl, Sakib Matin, Brendan Gifford, Galen T. Craven, Nicholas Lubbers

TL;DR

This work uses a graph-convolutional neural network architecture, the Hierarchically Interacting Particle Neural Network with Tensor Sensitivity (HIP-NN-TS), to develop a highly automated training pipeline for coarse-grained force fields, which allows for studying the transferability of coarse-grained models based on the force-matching approach.

Abstract

Coarse-graining is a molecular modeling technique in which an atomistic system is represented in a simplified fashion that retains the most significant system features that contribute to a target output, while removing the degrees of freedom that are less relevant. This reduction in model complexity allows coarse-grained molecular simulations to reach increased spatial and temporal scales compared to corresponding all-atom models. A core challenge in coarse-graining is to construct a force field that represents the interactions in the new representation in a way that preserves the atomistic-level properties. Many approaches to building coarse-grained force fields have limited transferability between different thermodynamic conditions as a result of averaging over internal fluctuations at a specific thermodynamic state point. Here, we use a graph-convolutional neural network architecture, the Hierarchically Interacting Particle Neural Network with Tensor Sensitivity (HIP-NN-TS), to develop a highly automated training pipeline for coarse grained force fields which allows for studying the transferability of coarse-grained models based on the force-matching approach. We show that this approach not only yields highly accurate force fields, but also that these force fields are more transferable through a variety of thermodynamic conditions. These results illustrate the potential of machine learning techniques such as graph neural networks to improve the construction of transferable coarse-grained force fields.

Thermodynamic Transferability in Coarse-Grained Force Fields using Graph Neural Networks

TL;DR

This work uses a graph-convolutional neural network architecture, the Hierarchically Interacting Particle Neural Network with Tensor Sensitivity (HIP-NN-TS), to develop a highly automated training pipeline for coarse-grained force fields, which allows for studying the transferability of coarse-grained models based on the force-matching approach.

Abstract

Coarse-graining is a molecular modeling technique in which an atomistic system is represented in a simplified fashion that retains the most significant system features that contribute to a target output, while removing the degrees of freedom that are less relevant. This reduction in model complexity allows coarse-grained molecular simulations to reach increased spatial and temporal scales compared to corresponding all-atom models. A core challenge in coarse-graining is to construct a force field that represents the interactions in the new representation in a way that preserves the atomistic-level properties. Many approaches to building coarse-grained force fields have limited transferability between different thermodynamic conditions as a result of averaging over internal fluctuations at a specific thermodynamic state point. Here, we use a graph-convolutional neural network architecture, the Hierarchically Interacting Particle Neural Network with Tensor Sensitivity (HIP-NN-TS), to develop a highly automated training pipeline for coarse grained force fields which allows for studying the transferability of coarse-grained models based on the force-matching approach. We show that this approach not only yields highly accurate force fields, but also that these force fields are more transferable through a variety of thermodynamic conditions. These results illustrate the potential of machine learning techniques such as graph neural networks to improve the construction of transferable coarse-grained force fields.
Paper Structure (12 sections, 9 equations, 10 figures)

This paper contains 12 sections, 9 equations, 10 figures.

Table of Contents

  1. Introduction
  2. Methods
  3. Coarse-graining
  4. Model architecture
  5. Training
  6. Performance evaluation
  7. Results
  8. Methanol Comparison Study
  9. Multi-molecule Study
  10. Computational cost
  11. Conclusions
  12. Data Availability Statement

Figures (10)

  • Figure 1: An illustration of the workflow used to create and analyze the ML CG models
  • Figure 2: An illustration of the three types of tests performed on the ML CG method.
  • Figure 3: Subfigures (a) and (b) show a comparison of methanol RDFs generated using (1) a reference AA simulation, (2) the MS-CG technique, and (3) the single-state ML CG models. These RDFs were generated at (a) 200 K, density 0.77 g/cm$^3$ and (b) 400 K, density 0.77 g/cm$^3$. Subfigure (c) summarizes the corresponding results across 11 temperatures. Subfigure (d) shows the transferability of a single-state model for each method, each trained with 300 K data.
  • Figure 4: Comparison of methanol ADFs generated using (1) a reference AA simulation, (2) the MS-CG technique, and (3) the single-state ML CG models. Subfigures (a), (c), and (e) show ADFs for 200 K, density 0.77 g/cm$^3$ with ADF cutoff values $r_\text{max}$ of 4 Å, 5 Å, and 6 Å respectively. Subfigures (b), (d), and (f) show ADFs for 400 K, density 0.77 g/cm$^3$ with the same respective cutoff values.
  • Figure 5: Summary of ADF comparisons for the MS-CG technique and the ML CG technique against a reference AA simulation. Subfigures (a), (c), and (e) show results for the single-state baseline models using cutoff value $r_\text{max}$ of (a) 4 Å, (c) 5 Å, and (e) 6 Å. Subfigures (b), (d), and (f) show results for the single-state transferability test using the 300 K model for each method with ADF cutoff values $r_\text{max}$ of (b) 4 Å, (d) 5 Å, and (f) 6 Å.
  • ...and 5 more figures