OpenMM: Scalable biomolecular modeling, simulation, and machine learning

Stanford University, Stanford CA

Investigators

Linked publications & trials

Abstract

PROJECT SUMMARY / ABSTRACT OpenMM is a widely used toolkit for molecular simulation and modeling (>1.5 million downloads, >2400 citations, >1M deployed instances). Its Python API makes it widely popular as both an application (for modelers) and a library (for developers), while its C/C++/Fortran bindings enable major legacy simulation packages to use OpenMM to provide high performance on modern hardware. OpenMM has been used for probing biological questions that leverage the $20B global investment in structural data from the PDB at multiple scales, from detailed studies of single disease proteins to superfamily-wide modeling studies and large-scale drug development efforts in industry and academia. In addition to having a very large user base, OpenMM is an essential component of many high-impact projects, such as AlphaFold2 and Folding@home. The first period of this grant enabled OpenMM to transition toward a community governance and sustainable development model, build a highly popular quantum chemical dataset that has enabled the community to build accurate machine learning (ML) potentials that deliver QM-level accuracy, and integrate support for fast GPU-accelerated protein:ligand simulations that make use of these ML potentials but can still achieve fast near-molecular mechanics (MM) speeds. In the next period, we propose to build on these successes to generate much more extensive datasets, integrate multiple other ML potentials to greatly expand the application domain they can be used for, and further accelerate and parallelize these simulations. We also aim to integrate key methodological improvements to improve accuracy for drug discovery applications. To be maximally responsive to the continually growing needs of the large OpenMM user community, we also aim to onboard a new developer focused on implementing community-requested features. To ensure OpenMM remains highly relevant to the modern machine learning revolution, we will continue to expand the ability of OpenMM to interoperate with modern ML frameworks such as TensorFlow, PyTorch, and JAX.