Implementation of Dual Resolution Simulation Methodology in LAMMPS, leading to improved studies of biological processes

eCSE04-07

Key Personnel

PI/Co-I: Jonathan Essex - University of Southampton, and Oliver Henrich - University of Edinburgh

Technical: Iain Bethune - University of Edinburgh, Sam Genheden - University of Southampton, and Sophia Wheeler - University of Southampton

Relevant documents

eCSE Technical Report: Implementation of Dual Resolution Simulation Methodology in LAMMPS

Project summary

Classical Molecular Dynamics simulations are widely used to understand the behaviour of biological systems. Usually the individual atoms are modelled explicitly in the calculations, but this is computationally demanding. A solution to this problem is to subsume groups of atoms into single particles or beads, thereby reducing the number of interaction sites. Such coarse-grain (CG) models are also very common, but inevitably simplifying the system loses accuracy. One possible solution is to use a hybrid model whereby the most critical parts of the system are represented at the atomistic level, with the remainder by coarse-graining. In this project, the simulation methods to allow such hybrid simulations have been updated and optimised in the LAMMPS simulation software.

Underlying any molecular simulation software is the integrator, which effectively solves Newton’s equations of motion in a series of short, discrete timesteps. Owing to the particular form of the CG model, our integrator needs to be able to solve the equations of motion associated with rotational motion efficiently. We implemented a new integrator in LAMMPS which is not only more accurate than that which was previously available, but also more efficient, allowing longer timesteps to be used while still conserving energy. Modifications to the pressure control algorithms were also implemented, together with improvements to the underlying code to increase performance of the software when runing on multiple processors at the same time. Taken together, these developments allow hybrid simulations of combined atomistic/coarse grain systems to be performed much more efficiently and accurately than was hitherto possible

This new functionality in LAMMPS is currently being used to tackle a range of scientific problems. First, we are exploring the effect of cholesterol on lipid bilayers. Cholesterol is an integral component of biological membranes, where it has a significant effect on lipid order and the mechanical flexibility of the bilayer. We are finding that a full coarse-grain model is not able to reproduce the so-called condensing effect of cholesterol, while our hybrid approach, in which cholesterol is modelled atomistically, does. Since we now know that our hybrid model is reliable, we are using it to simulate the effect of cholesterol on a particular class of membrane bound protein called a G-Protein Coupled Receptor (GPCR). GPCRs are intimately involved in cell signalling, and are a very important target for drug development. GPCRs are intrinsically highly flexible, and there is experimental evidence to suggest that the cholesterol composition of the membrane in which the GPCR sits is important in regulating GPCR shape, and hence biological activity. We are now simulating a number of GPCRs with various concentrations of cholesterol in its membrane, to explore this effect.

In addition, we are using the newly implemented algorithms to explore the phase transition behaviour of lipid systems, which again is an important factor in their biological function. We are also building a coarse-grain model for sugars, which will allow the hybrid simulation of glycoproteins i.e. proteins with a sugar coat. This important class of protein has hitherto been largely ignored by the simulation community owing to its complexity and a lack of efficient simulation models. Our hybrid approaches should prove very useful in this context.

These simulations will improve our understanding of the molecular-level behaviour of biological systems, and have the possibility of impacting on drug-development programmes.

Achievement of objectives

O1: The implementation of the symplectic and time-reversible rigid body integrator developed by Dullweber, Leimkuhler and McLachlan (DLM) in the LAMMPS software.

This integrator is added to LAMMPS as an extension of the fix nve/sphere code, enabled by the keywords update dipole/dlm. The original integrator is also retained as an option.

O2: The modification of the existing Parrinello Rahman constant pressure barostat to support the new integrator in the context of combined atomistic and rigid body molecular dynamics in LAMMPS.

The DLM integrator has been implemented in the FixNH (Nose-Hoover) base class, enabling its use for simulations in the NVT, NPT and NPH ensembles.

O3: The combination and optimisation of rRESPA multiple-timestep algorithm with this integrator and barostat in the context of dual-resolution molecular dynamics simulations in LAMMPS

Dual-resolution simulations using the new integrator and barostat have been tested. A new load balancing algorithm based on assigning weight factors to particles to account for the RESPA level (and hence frequency of force evaluations) has been implemented. Using a realistic system (atomistic BPTI in coarse-grained water), speed-ups of up to 33% have been demonstrated on 24 cores, and up to 103% on 72 cores.

O4: The modifications implemented in LAMMPS will be tested on a range of simulation systems of increasing complexity appropriate for determining the correctness and performance of the code.

In addition to the BPTI and pure water test cases described in the technical report, the Essex group have tested lipid membrane systems, and work is underway on extended simulations of lipid phase-change using metadynamics with PLUMED.

O5: All software to be demonstrated for its correctness and to be made available on ARCHER for the ARCHER community

The code developed for O1 and O2 is available in the latest LAMMPS stable release (30 Jul 2016). The code developed for O3 is available in the LAMMPS patch release (27 Sep 2016). A module lammps/elba has been installed on ARCHER which contains all the new code and is being used by the Essex group for extended testing.

Summary of the Software

The main LAMMPS website, with links to officially released versions, is http://lammps.sandia.gov - the latest stable release version (30 Jul 2016) already contains the new integrators developed in this project. Development versions of LAMMPS are kept in a github repository: https://github.com/lammps/lammps, in particular the lammps-icms branch is the ‘upstream’ version of LAMMPS, and contains features which will be rolled into the next stable release. All of the code changes developed in this project have been merged into lammps-icms, and have been released in the 27 Sep 2016 patch release on the LAMMPS website.

Various versions of LAMMPS are installed on ARCHER, and are freely available to all users (including source code, licensed under the GPL). Since LAMMPS may be configured with a wide range of optional packages, a version specialised for simulations with the ELBA force-field has been installed. This is based on the lammps-icms branch, with the addition of the PLUMED library for metadynamics. This module is called lammps/elba and is available to all users.

Development versions of the code developed during the eCSE may be found at https://github.com/ibethune/lammps, however everything has now been merged into upstream.