Graphein

Graphein constructs graph-based representations of protein structures and biological interaction networks for geometric deep learning and network analysis.

Key Features:

• Graph-Based Representations: Enables creation of graph models for protein structures and biomolecular interaction networks for use in geometric deep learning.
• Compatibility with Deep Learning Libraries: Produces graph objects compatible with DGL, PyTorch Geometric, and PyTorch3D.
• Framework Agnosticism: Built on the PyData ecosystem to interoperate with scientific computing libraries.
• Structural Data Retrieval: Provides utilities to retrieve structural data from the Protein Data Bank and the AlphaFold Structure Database.
• Interaction Network Retrieval: Provides utilities to retrieve biomolecular interaction networks from STRINGdb, BioGrid, TRRUST, and RegNetwork.
• Pre-processing Utilities: Includes pre-processing utilities to prepare experimental files and machine-learning-ready inputs.
• Network and Topological Analysis: Supports network-based, graph-theoretic, and topological analyses of structural and interaction datasets.
• High-Throughput Processing: Designed for high-throughput processing of large protein complexes and extensive interaction graphs.
• Dataset Provision: Provides protein structure-related datasets for geometric deep learning research.

Scientific Applications:

• High-Throughput Data Preparation: Processing large protein complexes and extensive interaction graphs into machine-learning-ready formats.
• Pre-processing for Experimental Files: Preparing experimental structural and interaction files for downstream analysis.
• Network-Based Analysis: Performing network-based, graph-theoretic, and topological analyses of structural and interaction datasets.
• Graph Representation Learning: Enabling graph representation learning on protein structures and biological networks.
• Drug Discovery: Supporting structural and network analyses relevant to drug discovery research.

Methodology:

Transforms raw structural and interaction data retrieved from the Protein Data Bank, the AlphaFold Structure Database, STRINGdb, BioGrid, TRRUST, and RegNetwork into machine-learning-ready graph datasets, exports graph objects compatible with DGL, PyTorch Geometric, and PyTorch3D, and includes pre-processing utilities.