Open Source Software for Transparent, Reproducible, Usable by Others, and Extensible High-Throughput Molecular Simulations

Abstract

Reproducibility and validation of scientific work by others is a key tenet of the scientific process and is absolutely necessary to increase confidence in the validity of said work. In practice, however, it can be quite challenging to reproduce the results of others’ work, especially in the field of molecular simulation. There are many steps in a molecular simulation workflow which can lead to irreproducibility, but a major source of these issues can be discovered and rectified during the preparation of the system of interest. Improving the reproducibility of the system preparation step is a necessary step forward as molecular simulations continue to increase in complexity and scale.

In order to improve the reproducibility of molecular simulations, major open-source software development efforts are required. This dissertation showcases the Molecular Simulation Design Framework (MoSDeF), a suite of Python libraries focused on improving the reproducibility and automatability of system preparation for molecular simulation. First, principals for more reproducible computational science and how MoSDeF enables this are examined. Two scientific applications using MoSDeF are then explored. First the property-structure relationship of coarse-grained patchy silica nanoparticles and the correlations between single nanoparticle properties and their self-assembled bulk phases are investigated. Finally, a large scale (~13000 unique systems) screening study of polymer coated silica substrates and the effect of terminal group chemistry on its tribological properties are explored.