EnzyNet: enzyme classification using 3D convolutional neural networks on spatial representation

EnzyNet is a project that uses 3D convolutional neural networks for enzyme classification. This webpage aims at detailing how to run and customize EnzyNet on your computer. We tried to make this package as user-friendly as possible, so that you can run tests on your own datasets too!

Also, if you are interested in the background of this project, you can take a look at our paper as well as our previous publications.

Installation

Compatibility

This code has been designed to work correctly with Python 3 on all platforms.

Procedure

To enjoy the features of this package, you will first have to download the project and go to the root of the folder.

cd /path/to/enzynet/

Then, just type

pip3 install -r requirements.txt

to install the dependencies.

Once this is done, we advise you to install the package in editable mode, so that you can modify the code as you wish and play with it right away. To do so, just type

pip3 install -e .

This is it, you are now ready to use EnzyNet!

Usage

Training

In order to train one of the 3D convolutional networks described in the paper, go to scripts/architecture/ and open the adequate Python file. Now, be sure to select

mode_run = 'train'

before running the script with the command

python3 enzynet_[weights-mode].py

where [weights-mode] corresponds to either uniform or adapted.

Testing

For testing phase, go to the same Python script as for training. This time, select

mode_run = 'test'

After you enter

python3 enzynet_[weights-mode].py

in the terminal, Python will load the weights that had been previously computed and stored in scripts/architecture/checkpoints/ in order to perform the predictions on the testing set.

Data visualization

In this package, we also developed a tool aimed at visualizing the PDB files in the representation that we adopted in the paper.

Binary	Hydropathy	Charge

For example, if you wish to see how the binary representation of enzyme 2Q3Z looks like in a cube of edge size 32, simply go to enzynet/visualization.py and fill out the script accordingly. In this case, it will be

if __name__ == '__main__':
  visualize_pdb('2Q3Z', vsize = 32, weights = None)

Then, just type

python3 visualization.py

in the terminal console to automatically compute and save the result in a PDF file to the folder scripts/volume/.

Customization

Dataset

If you want to do your experiments on another dataset, you have to make sure that it has the same structure as in datasets/partition_single.csv . If you don't, that's not a big deal! Just have a list of your PDB IDs in .txt files, like those in datasets/make_files/raw/. Then, you can perform the conversion by running the proper segments of the code in datasets/make_files/generate_datasets.py.

Architecture

Thanks to the user-friendliness of the high-level deep learning framework used in this project, you can also let your imagination run wild and very easily come up with your own architecture. To do so, you can go to scripts/architecture/ and directly modify the existing Python files accordingly.

Reference

You can find our paper on arXiv. The associated BibTeX code is

@article{Amidi2017,
author = {A. Amidi and S. Amidi and D. Vlachakis and V. Megalooikonomou and N. Paragios and E. Zacharaki},
title = {EnzyNet: enzyme classification using 3D convolutional neural networks on spatial representation},
journal = {ArXiv e-prints},
archivePrefix = {arXiv},
eprint = {1707.06017}
}

Acknowledgements

This project relies on the great deep learning framework Keras coupled with Google's open-source TensorFlow, as well as on the very well documented PDB parser BioPython.