Learning protein sequence embeddings using information from structure

New and improved embedding models combining sequence and structure training are now available at https://github.com/tbepler/prose!

<br /> <br />

This repository contains the source code and links to the data and pretrained embedding models accompanying the ICLR 2019 paper: Learning protein sequence embeddings using information from structure

@inproceedings{
bepler2018learning,
title={Learning protein sequence embeddings using information from structure},
author={Tristan Bepler and Bonnie Berger},
booktitle={International Conference on Learning Representations},
year={2019},
}

Setup and dependencies

Dependencies:

• python 3
• pytorch >= 0.4
• numpy
• scipy
• pandas
• sklearn
• cython
• h5py (for embedding script)

Run setup.py to compile the cython files:

python setup.py build_ext --inplace

Data sets

The data sets with train/dev/test splits are provided as .tar.gz files from the links below.

• SCOPe data
• Pfam data
• Protein secondary structure data
• Transmembrane data
• CASP12 contact map data

The training and evaluation scripts assume that these data sets have been extracted into a directory called 'data'.

Pretrained models

Our trained versions of the structure-based embedding models and the bidirectional language model can be downloaded here.

Author

Tristan Bepler (tbepler@mit.edu)

Cite

Please cite the above paper if you use this code or pretrained models in your work.

License

The source code and trained models are provided free for non-commercial use under the terms of the CC BY-NC 4.0 license. See LICENSE file and/or https://creativecommons.org/licenses/by-nc/4.0/legalcode for more information.

Contact

If you have any questions, comments, or would like to report a bug, please file a Github issue or contact me at tbepler@mit.edu.