BioSimSpace: An interoperable Python framework for biomolecular simulation

Abstract

Biomolecular simulation is a diverse and growing area of research, making important contributions to structural biology and pharmaceutical research (Huggins et al., 2019). Within the community there are a several significant and widely used software packages that have evolved from within various research groups over the past 20 or more years. For example, the molecular dynamics packages AMBER (Case et al., 2005), GROMACS (Abraham et al., 2015), and NAMD (Phillips et al., 2005), which are all capable of running biomolecular simulations for systems consisting of hundreds of thousands of atoms and can be run on hardware ranging from laptops, to graphics processing units (GPUs), to the latest high-performance computing clusters. Since different software packages were developed independently, interoperability between them is poor. In large part this is the result of major differences in the supported file formats, which makes it difficult to translate the inputs and outputs of one program to another. As a consequence, expertise in one package doesn't immediately apply to another, making it hard to share methodology and knowledge between different research communities , as evidenced, for instance, by a recent study on reproducibility of relative hydration free energies across simulation packages (Loeffler et al., 2018). The issue is compounded by the increasing use of biomolecular simulations as components of larger scientific workflows for bio-engineering or computer-aided drug design purposes. A lack of interoperability leads to brittle workflows, poor reproducibility, and lock in to specific software that hinders dissemination of biomolecular simulation methodologies to other communities. Several existing software packages attempt to address this problem: InterMol (Shirts et al., 2016) and ParmEd (Swails, Jason, 2010) can be used to read and write a wide variety of common molecular file formats; ACPYPE (Sousa da Silva & Vranken, 2012) can generate small molecule topologies and parameters for a variety of molecular dynamics engines; MDTraj (McGibbon et al., 2015) and MDAnalysis (Gowers et al., 2016) support reading, writing, and analysis of different molecular trajectory formats; the Atomic Simulation Engine (ASE) handles a wide variety of atomistic simulation tasks and provides interfaces to a range of external packages; and the Cuby (Řezáč, 2016) framework allows access to a range of computational chemistry functionality from external packages, which can be combined into complex workflows through structured input files. Despite their utility, the above packages either have a restricted domain of application, e.g. trajectory files, or require different configuration options or scripts to interface with different external packages. It is not possible to write a single script that is independent of the underlying software packages installed on the host system.