Abstract:Coarse-grained (CG) molecular dynamics enables the study of biological processes at temporal and spatial scales that would be intractable at an atomistic resolution. However, accurately learning a CG force field remains a challenge. In this work, we leverage connections between score-based generative models, force fields and molecular dynamics to learn a CG force field without requiring any force inputs during training. Specifically, we train a diffusion generative model on protein structures from molecular dynamics simulations, and we show that its score function approximates a force field that can directly be used to simulate CG molecular dynamics. While having a vastly simplified training setup compared to previous work, we demonstrate that our approach leads to improved performance across several small- to medium-sized protein simulations, reproducing the CG equilibrium distribution, and preserving dynamics of all-atom simulations such as protein folding events.

What problem does this paper attempt to address?

The paper aims to address a key challenge in Coarse-Grained (CG) molecular dynamics simulations: how to effectively learn a CG force field so that CG molecular dynamics simulations can reproduce important features of atomistic simulations, especially in achieving the study of biological processes on time and spatial scales. To solve this problem, the authors leverage the connection between Score-Based Generative Models, force fields, and molecular dynamics, proposing a new method to learn CG force fields without requiring any force information during the training process. Specifically, they achieve this goal through the following steps: 1. **Model Training**: The authors use a diffusion generative model to train on protein structures obtained from molecular dynamics simulations. 2. **Force Field Extraction**: By demonstrating the connection between score-based generative models, force fields, and molecular dynamics, the authors show that the score function of the generative model can approximate a force field, which can be directly used for CG molecular dynamics simulations. 3. **Performance Validation**: Compared to previous work, this method greatly simplifies the training setup and demonstrates improved performance in multiple protein simulations, including systems with up to 56 amino acids, being able to reproduce CG equilibrium distributions and maintain the dynamical mechanisms of all-atom simulations such as protein folding. This method not only provides a simple single-stage training framework but also shows excellent performance in several protein simulations, particularly demonstrating potential advantages in simulating larger proteins. Additionally, this method overcomes some limitations of traditional methods, such as low data efficiency, high computational cost, and insufficient model accuracy.

Two for One: Diffusion Models and Force Fields for Coarse-Grained Molecular Dynamics

Learning Effective Molecular Models from Experimental Observables.

Machine Learning of coarse-grained Molecular Dynamics Force Fields

Statistically Optimal Force Aggregation for Coarse-Graining Molecular Dynamics

Top-down machine learning of coarse-grained protein force-fields

Navigating protein landscapes with a machine-learned transferable coarse-grained model

Learning data efficient coarse-grained molecular dynamics from forces and noise

Simulate Time-integrated Coarse-grained Molecular Dynamics with Multi-Scale Graph Networks

Ensemble learning of coarse-grained molecular dynamics force fields with a kernel approach.

Diffusion-Driven Generative Framework for Molecular Conformation Prediction

Coarse Graining Molecular Dynamics with Graph Neural Networks

DiffMD: A Geometric Diffusion Model for Molecular Dynamics Simulations

A Scalable Graph Neural Network Method For Developing An Accurate Force Field Of Large Flexible Organic Molecules

Learning coarse-grained force fields for fibrogenesis modeling

Modeling Molecular Structures with Intrinsic Diffusion Models

Multi-body Effects in a Coarse-Grained Protein Force Field.

Coarse-Grained Molecular Dynamics Study Based on TorchMD

Development of Coarse-Grained Models for Polymers by Trajectory Matching

Consistent View of Protein Fluctuations from All-Atom Molecular Dynamics and Coarse-Grained Dynamics with Knowledge-Based Force-Field.

Transferable Coarse Graining Via Contrastive Learning of Graph Neural Networks

Towards Exact Molecular Dynamics Simulations with Machine-Learned Force Fields