Documentation | Paper

Protein Graph Library

This package provides functionality for producing a number of types of graph-based representations of proteins. We provide compatibility with standard formats, as well as graph objects designed for ease of use with popular deep learning libraries.

Example usage

from graphein.construct_graphs import  ProteinGraph
from graphein.construct_meshes import  ProteinMesh

# Initialise ProteinGraph class
pg = ProteinGraph(granularity='CA', insertions=False, keep_hets=True,
                  node_featuriser='meiler', get_contacts_path='/Users/arianjamasb/github/getcontacts',
                  pdb_dir='examples/pdbs/',
                  contacts_dir='examples/contacts/',
                  exclude_waters=True, covalent_bonds=False, include_ss=True)

# Create residue-level graphs. Chain selection is either 'all' or a list e.g. ['A', 'B', 'D'] specifying the polypeptide chains to capture

# DGLGraph From PDB Accession Number
graph = pg.dgl_graph_from_pdb_code('3eiy', chain_selection='all')
# DGLGraph From PDB file
graph = pg.dgl_graph_from_pdb_file('examples/pdbs/3eiy.pdb', chain_selection='all')

# Create atom-level graphs
graph = pg.make_atom_graph(pdb_code='3eiy', graph_type='bigraph')

# Initialise ProteinMesh class
pm = ProteinMesh()

# Pytorch3D Mesh Object from PDB Code
verts, faces, aux = pm.create_mesh(pdb_code='3eiy', out_dir='examples/meshes/')
# Pytorch3D Mesh Object from PDB File
verts, faces, aux = pm.create_mesh(pdb_file='examples/pdbs/pdb3eiy.pdb')

Parameters

Graphs can be constructed according to walks through the graph in the figure below.

granularity: {'CA', 'CB', 'atom'} - specifies node-level granularity of graph
insertions: bool - keep atoms with multiple insertion positions
keep_hets: bool - keep hetatoms
node_featuriser: {'meiler', 'kidera'} low-dimensional embeddings of AA physico-chemical properties
pdb_dir: path to pdb files
contacts_dir: path to contact files generated by get_contacts
get_contacts_path: path to GetContacts installation
exclude_waters: bool - retain structural waters
covalent_bonds: bool - maintain covalent bond edges or just use intramolecular interactions
include_ss: bool - calculate protein SS and surface features using DSSP and assign them as node features

Installation

1. Create env

   conda create --name graphein python=3.7
   conda activate graphein

2. Install Get Contacts

    # Install get_contact_ticc.py dependencies
    $ conda install scipy numpy scikit-learn matplotlib pandas cython seaborn
    $ pip install ticc==0.1.4
     
    # Install vmd-python dependencies
    $ conda install netcdf4 numpy pandas seaborn  expat tk=8.5  # Alternatively use pip
    $ brew install netcdf pyqt # Assumes https://brew.sh/ is installed
   
    # Install vmd-python library
    $ conda install vmd-python
   
    # Set up getcontacts library
    $ git clone https://github.com/getcontacts/getcontacts.git
    $ echo "export PATH=`pwd`/getcontacts:\$PATH" >> ~/.bash_profile
    $ source ~/.bash_profile
   
    # Test installation
    $ cd getcontacts/example/5xnd
    $ get_dynamic_contacts.py --topology 5xnd_topology.pdb \
                              --trajectory 5xnd_trajectory.dcd \
                              --itypes hb \
                              --output 5xnd_hbonds.tsv

3. Install Biopython

   conda install biopython

4. Install RDKit

   conda install -c conda-forge rdkit==2018.09.3

5. Install DSSP

   We use DSSP for computing some protein features

   $ conda install -c salilab dssp

6. Install PyTorch Geometric

   $ pip install torch-scatter
   $ pip install torch-sparse
   $ pip install torch-spline-conv
   $ pip install torch-cluster
   $ pip install torch-geometric

   N.B. Follow the instructions in the Torch-Geometric Docs to install the versions appropriate to your CUDA version.

7. Install IPyMol

   $ git clone https://github.com/cxhernandez/ipymol
   $ cd ipymol
   $ pip install . 

   N.B. The PyPi package seems to be behind the github repo. We require functionality that is not present in the PyPi package.

8. Install DGL and associated libraries

   $ conda install -c dglteam dgl
   $ conda install -c dglteam dgllife

9. Install graphein

   $ git clone https://www.github.com/a-r-j/graphein
   $ cd graphein
   $ pip install -e .

Citing Graphein

Please consider citing graphein if it proves useful in your work.

@article{Jamasb2020,
  doi = {10.1101/2020.07.15.204701},
  url = {https://doi.org/10.1101/2020.07.15.204701},
  year = {2020},
  month = jul,
  publisher = {Cold Spring Harbor Laboratory},
  author = {Arian Rokkum Jamasb and Pietro Lio and Tom Blundell},
  title = {Graphein - a Python Library for Geometric Deep Learning and Network Analysis on Protein Structures}
}