Automatic learning of hydrogen-bond fixes in an AMBER RNA force field

In this paper we used a technique to fit force field parameters on data bases of molecular simulations trajectories. The work also required the production of an enormous amount of simulation data, which cannot be reproduced in a short time. However, we uploaded all the preprocessed trajectories on Zenodo, so that it should be possible to perform the training again in a few days.

All the analysis was done in Python using Jupyter Notebooks that are included in the linked repository. In addition to standard Python packages, one would have to install our group tools (https://github.com/bussilab/py-bussilab), to perform weighted histogram analysis, and the cudamat library (https://github.com/cudamat/cudamat), to run the analysis using GPUs.

We do care about reproducibility. In case we receive any feedback, we would be really happy to improve our Github repository and/or submitted manuscript so as to make the reproduction easier!

  Paper URL: https://arxiv.org/abs/2201.04078

  Data URL: https://zenodo.org/record/5851199

  Code URL: https://github.com/bussilab/ghbfix-training

Tags: Python HPC machine learning Molecular Dynamics

HPC ReproHack @ Warwick