OpenMM 8: Molecular Dynamics Simulation with Machine Learning Potentials

Eastman P, Galvelis R, Peláez RP, Abreu CRA, Farr SE, Gallicchio E, Gorenko A, Henry MH, Hu F, Huang J, Krämer A, Michel J, Mitchell J, Pande VS, Rodrigues JPGLM, Rodriguez-Guerra J, Simmonett AC, Swails J, Turner P, Wang Y, Zhang I, Chodera JD, De Fabritiis G, Markland TE
Journal of Physical Chemistry B [DOI] [website] [code]

We present OpenMM 8, which includes GPU-accelerated support for simulating hybrid ML/MM systems that use machine learning (ML) potentials to achieve high accuracy with minimal loss in speed.

Machine-learned molecular mechanics force field for the simulation of protein-ligand systems and beyond

Takaba K, Pulido I, Behara PK, Henry M, MacDermott-Opeskin H, Chodera JD, Wang Y
preprint: [arXiv]

We present a new self-consistent MM force field trained on $>$1.1M quantum chemical calculations that uses graph nets to achieve high accuracy and produce accurate protein-ligand binding free energies.

EspalomaCharge: Machine learning-enabled ultra-fast partial charge assignment

Wang Y, Pulido I, Takaba K, Kaminow B, Scheen J, Wang L, Chodera JD
preprint: [arXiv]

We present a drop-in replacement for generating AM1-BCC ELF10 charges based on graph convolutional nets that is orders of magnitude faster than standard methods for both small molecules and biomolecules.

Spatial attention kinetic network with E(n) equivariance

Yuanqing Wang and John D. Chodera
preprint: [arXiv] [code]

This work descibes Spatial Attention Kinetic Networks (SAKE), a new E(n)-equivariant architecture that uses spatial attention, enabling the construction of extremely performant but still accurate machine learning potentials, as well as flows capable of prediction dynamics.

SPICE, A Dataset of Drug-like Molecules and Peptides for Training Machine Learning Potentials

Eastman P, Behara PK, Dotson DL, Galvelis R, Herr JE, Horton JT, Mao Y, Chodera JD, Pritchard BP, Wang Y, De Fabritiis G, and Markland TE
Scientific Data 10:11, 2023 [DOI]

To remedy the lack of large, open quantum chemical datasets for training accurate general machine learning potentials and molecular mechanics force fields for druglike small molecules and biomolecules, we produce the open SPICE dataset, and show how it can be used to build extremely accurate machine learning potentials.

End-to-end differentiable molecular mechanics force field construction

Yuanqing Wang, Josh Fass, and John D. Chodera
Chemical Science 13:12016, 2022 [DOI] [arXiv] [pytorch code] [JAX code]

Molecular mechanics force fields have been a workhorse for computational chemistry and drug discovery. Here, we propose a new approach to force field parameterization in which graph convolutional networks are used to perceive chemical environments and assign molecular mechanics (MM) force field parameters. The entire process of chemical perception and parameter assignment is differentiable end-to-end with respect to model parameters, allowing new force fields to be easily constructed from MM or QM force fields, extended, and applied to arbitrary biomolecules.

Graph nets for partial charge prediction

Yuanqing Wang, Josh Fass, Chaya D. Stern, Kun Luo, and John D. Chodera.
Preprint ahead of publication. [arXiv] [GitHub]

Graph convolutional and message-passing networks can be a powerful tool for predicting physical properties of small molecules when coupled to a simple physical model that encodes the relevant invariances. Here, we show the ability of graph nets to predict partial atomic charges for use in molecular dynamics simulations and physical docking.