OpenAWSEM with Open3SPN2: A fast, flexible, and accessible framework for large-scale coarse-grained biomolecular simulations

Wei Lu, Carlos Bueno, Nicholas P. Schafer, Joshua Moller, Shikai Jin, Xun Chen, Mingchen Chen, Xinyu Gu, Aram Davtyan, Juan J. de Pablo, Peter G. Wolynes

Introduction

Simulating large-scale biomolecular complexes, such as the ribosome or CRISPR-Cas9, presents a significant "timescale gap." While all-atom simulations provide high detail, they are computationally prohibitive for the long durations required to observe complex cellular processes. Coarse-grained (CG) models, which group atoms into "beads," offer a solution by simplifying the physics. However, traditional implementations in packages like LAMMPS often struggle with efficiency on modern hardware when dealing with implicit solvent systems.

This paper introduces OpenAWSEM and Open3SPN2, a framework that ports the AWSEM (protein) and 3SPN2 (DNA) models to the GPU-accelerated OpenMM environment. This transition provides a 30- to 300-fold speedup while making the models significantly easier for researchers to customize and extend.

Methods and Implementation

The researchers leveraged OpenMM, a library optimized for high-performance molecular dynamics on GPUs. A key feature of this implementation is the use of OpenMM’s Custom Forces framework. This allows researchers to define complex energy terms using simple mathematical expressions in Python, which the software then automatically compiles into optimized GPU code. This "hackable" approach removes the need for deep expertise in C++ or CUDA to add new physics.

Two sophisticated features were implemented using this approach:

  1. Z-dependent Membrane Potential: A potential that dynamically switches between "water-like" and "lipid-like" interactions based on a residue's vertical position relative to a cell membrane, trained on 1,561 proteins from the OPM database.
  2. Multi-body Disulfide Bonding: A novel term that uses a density variable to ensure a cysteine residue only forms one bond at a time (saturation), preventing the unphysical clustering seen in simpler pairwise models.

Key Insights: Force-based Parallelization

A major insight of this work is the superiority of force-based parallelization over spatial parallelization for implicit solvent models. Traditional spatial decomposition (used in LAMMPS) is efficient for bulk liquids but leaves many processors idle when simulating a single protein in a large, mathematically simplified solvent box. By parallelizing the calculation of all interactions across the GPU, OpenAWSEM maintains high efficiency regardless of the system's spatial distribution.

Results and Performance

The OpenMM-based framework demonstrates dramatic performance improvements over CPU-bound LAMMPS implementations across various system sizes.

Performance comparison of OpenMM vs. LAMMPS for protein simulations of varying lengths.

As shown in the figure above, the time required for protein simulations scales much more favorably in OpenMM. For large proteins, OpenAWSEM achieved a 30-fold speedup. For specific protein-DNA systems like RecA, the speedup reached 300-fold.

Scaling of simulation time with DNA length for LAMMPS and OpenMM.

Beyond raw speed, the framework improved topological assignments for membrane proteins and correctly modeled disulfide bond formation in Ribonuclease A, replicating classic experimental results that simpler models fail to capture.

Conclusion and Future Directions

OpenAWSEM and Open3SPN2 provide a powerful, accessible tool for the biophysics community. By using Python 3 for the user interface, the authors have lowered the barrier to entry for modifying and extending these sophisticated models.

Future work will focus on refining specific chemical interactions at the protein-DNA interface and scaling the framework to "cellular-scale" simulations involving hundreds of interacting molecules. The ability to simulate massive biomolecular complexes on timescales approaching actual cellular processes represents a significant step forward for computational biology.

Summary updated

Abstract

Abstract We present OpenAWSEM and Open3SPN2, new cross-compatible implementations of coarse-grained models for protein (AWSEM) and DNA (3SPN2) molecular dynamics simulations within the OpenMM framework. These new implementations retain the chemical accuracy and intrinsic efficiency of the original models while adding GPU acceleration and the ease of forcefield modification provided by OpenMM’s Custom Forces software framework. By utilizing GPUs, we achieve more than a 100-fold speedup in protein and protein-DNA simulations over the existing LAMMPS-based implementations running on a CPU. We showcase the benefits of OpenMM’s Custom Forces framework by devising and implementing two new potentials that allow us to address important aspects of protein folding and structure prediction and by testing the ability of the combined OpenAWSEM and Open3SPN2 to model protein-DNA binding. The first potential is used to describe the changes in effective interactions that occur as a protein becomes partially buried in a membrane. We also introduced an interaction to describe proteins with multiple disulfide bonds. Using simple pairwise disulfide bonding terms results in unphysical clustering of cysteine residues, posing a problem when simulating the folding of proteins with many cysteines. We now can computationally reproduce Anfinsen’s early Nobel prize winning experiments [1] by using OpenMM’s Custom Forces framework to introduce a multi-body disulfide bonding term that prevents unphysical clustering. Our protein-DNA simulations show that the binding landscape is funneled towards structures that are quite similar to those found using experiments. In summary, this paper provides a simulation tool for the molecular biophysics community that is both easy to use and sufficiently efficient to simulate large proteins and large protein-DNA systems that are central to many cellular processes. These codes should facilitate the interplay between molecular simulations and cellular studies, which have been hampered by the large mismatch between the time and length scales accessible to molecular simulations and those relevant to cell biology. Author summary The cell’s most important pieces of machinery are large complexes of proteins often along with nucleic acids. From the ribosome, to CRISPR-Cas9, to transcription factors and DNA-wrangling proteins like the SMC-Kleisins, these complexes allow organisms to replicate and enable cells to respond to environmental cues. Computer simulation is a key technology that can be used to connect physical theories with biological reality. Unfortunately, the time and length scales accessible to molecular simulation have not kept pace with our ambition to study the cell’s molecular factories. Many simulation codes also unfortunately remain effectively locked away from the user community who need to modify them as more of the underlying physics is learned. In this paper, we present OpenAWSEM and Open3SPN2, two new easy-to-use and easy to modify implementations of efficient and accurate coarse-grained protein and DNA simulation forcefields that can now be run hundreds of times faster than before, thereby making studies of large biomolecular machines more facile.
BibTeX
@article{lu2021openawsem,
  author = {Wei Lu and Carlos Bueno and Nicholas P. Schafer and Joshua Moller and Shikai Jin and Xun Chen and Mingchen Chen and Xinyu Gu and Aram Davtyan and Juan J. de Pablo and Peter G. Wolynes},
  title = {{OpenAWSEM with Open3SPN2: A fast, flexible, and accessible framework for large-scale coarse-grained biomolecular simulations}},
  journal = {PLOS Computational Biology},
  year = {2021},
  doi = {10.1371/journal.pcbi.1008308},
}
Figure