repository·master·Indexed 22 days ago
https://github.com/a-r-j/grapheinA specialized library for generating geometric graph representations of biological structures, including proteins, RNA, and molecules, for use in deep learning workflows. It features support for AlphaFold2 data, molecular graphs from SMILES, RNA graphs from dotbracket notation, and protein mesh generation. Graphein includes integrated PyTorch Geometric dataloaders and provides worked datasets for tasks such as Protein-Protein Interaction Site Prediction (PPISP) and the Protein Structural Change Database (PSCDB).
Graphein is a Python package designed to produce various graph-based representations of biomolecular structures. It provides compatibility with standard geometric deep learning library formats, allowing you to export graphs to:
NetworkX (nx.Graph)PyTorch Geometric (pytorch_geometric.data.Data)DGL (dgl.DGLGraph)It also provides specialized graph objects optimized for use with popular deep learning libraries.
Graphein provides several worked examples of creating graph datasets from protein structures. These datasets are designed for various machine learning tasks, including node classification (e.g., predicting interaction sites), edge prediction (e.g., predicting bound conformations), and graph classification (e.g., predicting structural rearrangement classes).
Key dataset categories include:
graphein.ppi module provides tools for constructing graphs representing Protein-Protein Interactions. You can build these graphs using high-level graph construction functions, define specific edges, and extract graph or node features. The module also supports parsing interaction data from databases like BioGrid and STRINGDB.Graphein has introduced several major capabilities across its versions:
Graphein interfaces with several popular geometric deep learning libraries to facilitate downstream machine learning tasks. Supported libraries include:
Edges in protein graphs can be defined using several different criteria through the graphein.protein.edges submodules:
graphein.protein.edges.distance).graphein.protein.edges.intramolecular).graphein.protein.edges.atomic).You can set up a complete development environment using the provided Conda YAML files. This is recommended if you need specific GPU builds (CUDA 11.1) for integrated deep learning libraries.
# For a full dev environment (includes GPU builds)
git clone https://www.github.com/a-r-j/graphein
cd graphein
conda env create -f environment-dev.yml
pip install -e .
# For a lighter install
git clone https://www.github.com/a-r-j/graphein
cd graphein
conda env create -f environment.yml
pip install -e .If you intend to contribute to Graphein or need a development environment that includes GPU builds (CUDA 11.1) for geometric deep learning libraries, use the following methods.
Pip installation:
pip install graphein[dev]pip install graphein[all]Conda installation:
environment-dev.yml.pip install -e ..A devcontainer is also provided for running the development environment in a lightweight container via VS Code.
# Pip
pip install graphein[dev] # For dev dependencies
pip install graphein[all] # To get the lot
# Conda
git clone https://www.github.com/a-r-j/graphein
cd graphein
conda env create -f environment-dev.yml
pip install -e .The PROTEINS_LIGANDS dataset contains protein-ligand interactions. It is a collection of non-redundant PDB chains of ligand-interacting protein chains.
Dataset Properties:
Available Subsets:
| PDB Ligand | Ligand Name | # Proteins | Dataset |
|---|---|---|---|
| SO4 | Sulphate | 3312 | SO4_3312 |
| PO4 | Phosphate | 1299 | PO4_1299 |
| NAG | N-Acetylglucosamine | 727 | NAG_727 |
| HEM | Heme | 176 | HEM176 |
| BME | Beta-Mercaptoethanol | 191 | BME191 |
| EDO | Ethylene Glycol | 1507 | EDO1507 |
| PLP | Vitamin B6 Phosphate | 65 | PLP65 |
For a full walkthrough on how to parse this dataset, use the provided Colab notebook.
https://colab.research.google.com/research/urls?urls=https://github.com/a-r-j/graphein/blob/master/datasets/proteins_ligands/parse_dataset.ipynbGraphein supports various deep learning configurations. While the Conda dev environment includes CUDA 11.1 and PyTorch, you can manually install these libraries via Conda:
pytorchpytorch3d (requires fvcore and iopath)dgl (via dglteam channel)pytorch-geometric (via rusty1s and conda-forge channels)conda install -c pytorch pytorch
conda install -c pytorch3d pytorch3d
conda install -c dglteam dgl
conda install pytorch-geometric -c rusty1s -c conda-forgeSome Graphein features require external utilities not available via PyPI.
Use the following commands to install them via Conda or manual setup:
conda install -c salilab dssp # Required for computing secondary structural features
conda install -c schrodinger pymol # Required for PyMol visualisations & mesh generation
# GetContacts setup
conda install -c conda-forge vmd-python
git clone https://github.com/getcontacts/getcontacts
echo "export PATH=$PATH:`pwd`/getcontacts" >> ~/.bashrc
source ~/.bashrc