HylleraasMD: Massively parallel hybrid particle-field molecular dynamics in Python

Abstract

Molecular dynamics (MD) is a computational methodology in which the dynamical behavior of systems of interacting atoms and molecules is investigated by integrating the corresponding classical equations of motion. The analysis of the molecular trajectories yields an incredibly powerful computational microscope with atomic resolution. While prominent examples of molecular dynamics involving all-atom models exist, many systems operate on time-and lengths scales too large, precluding the use of such an approach. The intrinsic complexity of biological soft-matter systems has necessitated the development of coarse-grained (CG) MD models wherein groups of atoms are treated as individual entities. To probe experimentally relevant length-(nm-µm) and time-(ps-ms) scales, further reduction of computational complexity may be warranted through the removal of explicit particle-particle interactions in favor of particle-density field interactions. Such hybrid particle-field (hPF) models recast the interactions between particle pairs into a system of free particles interacting with an external potential dependent on the density, in analogy with self-consistent field theories.HylleraasMD (named after our affiliate centre, the Hylleraas Centre for Quantum Molecular Sciences) (HyMD) is a Python package capable of highly parallel hPF-MD simulations of a wide range of surfactants and other biological systems in a CG representation. At present, it is the only open-source implementation of the hPF formalism freely available to computational researchers.

Soft Matter under Pressure: Pushing Particle–Field Molecular Dynamics to the Isobaric Ensemble

“…More details can be found in ref . Our HhPF approach was implemented into the HylleraasMD (HyMD) modeling package. − More on this and the implementation of the barostat can be found in SI: Pressure Implementation.…”

Learning Force Field Parameters from Differentiable Particle-Field Molecular Dynamics

“…3 can be achieved in multiple ways, such as in Monte Carlo, often referred to as the Single-Chain-In-Mean-Field method, 8 or by molecular dynamics with various formulations. 9,17,36,37 We adopt our recently developed Hamiltonian approach, which is the only implementation demonstrated to achieve energy-conserving and alias-free dynamics. 14 In brief, this approach builds on standard particle-mesh operations 38 by assigning particle number densities onto a regular grid via an assignment function P and subsequently performing a convolution with a filter function G that defines the density spread associated with particle species t as…”

“…While particle-field simulations are maturing both in terms of mathematical foundations 14,15 and open-source software implementation, 9,16,17 the parameterization of the particle-field force field remains challenging. On top of being coarse-grained, which targets the more elusive potential of mean force instead of the potential energy surface of all-atom simulations, the determination of the non-standard particle-field interactions is not straightforwardly amenable to traditional techniques, such as iterative Boltzmann inversion or force-matching.…”

SANS Spectra with PLUMED: Implementation and Application to Metainference

“…To test how metainference using the SANS intensities performs in the case of softer potentials, we simulated DPC micelles using the HhPF approach, as implemented in HylleraasMD (HyMD). , We use the same mapping for the surfactant as that used by the MARTINI force field. Simulations were performed in the NVT ensemble at 300 K using the CSVR thermostat, with one DPC micelle composed of 54 surfactants solvated in a cubic water box of ∼8 nm per side.…”

“…We showcase and validate the implementation using it with metainference through coarse-grained MD simulations for beta-octylglucoside (BOG) micelles and dodecylphosphocholine (DPC) micelles in water. These simulations were performed using two different software (GROMACS 22 and Hylleraas MD 23 ), and against two coarse-grained modeling approaches. In the case of BOG, we observe that the MARTINI model 24 greatly favors the formation of big micelles that do not correspond to the SANS intensities at concentrations close to the critical micelle concentration (cmc).…”