The Python programming language is easy to use, enables high-level code, and comes with an excellent ecosystem of libraries. It has been widely adopted in the scientific community for purposes ranging from "glue" code between different software packages and small scripts to large computational packages. In biophysics and biochemistry, classical molecular dynamics (MD) simulations are used to study the relationship between structure and function of biomolecules at the atomic level. Increasingly it is necessary to run whole ensembles of simulations to address specific questions in a reproducible and quantitative manner but it has become challenging to run hundreds of coupled simulations (possibly on heterogeneous computing resources) and to analyze them (trajectories with terabytes of data). We have been developing Python-based open source tools as building blocks for ad-hoc and automated workflows to generate and analyze MD trajectories. Our widely used MDAnalysis library (https://mdanalysis.org) provides an abstract, object-oriented interface to MD data in practically all common file formats together with common analysis building blocks. Analysis can be parallelized either with RADICAL-Pilot (https://radical-cybertools.github.io/radical-pilot/) or with our new Parallel MDAnalysis library (PMDA, https://mdanalysis.org/pmda/), which serves as a framework for MDAnalysis to use asynchronous task-level parallelism provided by Dask (https://dask.org/). PMDA can seamlessly make use of resources ranging from multi-core laptops to supercomputers. We have been using the FireWorks (https://materialsproject.github.io/fireworks/) workflow management library to prototype a workflow for absolute free energy calculations that runs on national XSEDE supercomputers, university supercomputers, and local workstations. The MD simulations are performed with Gromacs (http://www.gromacs.org) but the absence of native Python bindings necessitated the use of a thin Pythonic wrapper library (GromacsWrapper https://github.com/Becksteinlab/GromacsWrapper), which may serve as an example for how to quickly integrate other codes into Python-based workflows. By maintaining complete workflows within Python, it is easy to directly interface with other packages for scientific computations, time series analysis, plotting, graph analysis, and machine learning, and thus increase productivity and enable new ways to answer scientific questions.

Integration via Python: Building blocks for simulation and analysis workflows for molecular dynamics

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.