molearn

protein conformational spaces meet machine learning

molearn is a Python package streamlining the implementation of machine learning models dedicated to the generation of protein conformations from example data obtained via experiment or molecular simulation.

Included in this repository are the following:

Source code in the molearn folder
Software documentation in the docs folder.
Example training and analysis scripts, along with example data, in the examples folder

Dependencies

The current version of molearn only supports Linux, and has verified to support Python >=3.9.

Required Packages

numpy
PyTorch (1.7+)
Biobox

Optional Packages

To run energy evaluations with OpenMM:

To evaluate Sinkhorn distances during training:

geomloss

To calculate DOPE and Ramachandran scores during analysis:

Modeller (requires academic license)
cctbx

To run the GUI:

Installation

molearn requires no installation. Simply clone the repository and make sure the requirements above are met. The most recent release can also be obtained through Anaconda:

conda install molearn -c conda-forge

Usage

See example scripts in the examples folder.
Jupyter notebook tutorials describing the usage of a trained neural network are available here.
software documentation in the docs folder is available at molearn.readthedocs.io.

Reference

If you use molearn in your work, please cite: V.K. Ramaswamy, S.C. Musson, C.G. Willcocks, M.T. Degiacomi (2021). Learning protein conformational space with convolutions and latent interpolations, Physical Review X 11

Contact

For any question please contact samuel.musson@durham.ac.uk

DerienFe/molearn