GBPNet: Universal Geometric Representation Learning on Protein Structures

The structure of the project is as follows:

• gbpnet.models: The implementation of the models used in the experiments.
• gbpnet.datamodules: The data processing pipeline.
• run_test.py: The main script for reproducing the results.

Installation

The verified versions of the dependencies can be found in requirements.txt.

Data

The datasets are automatically downloaded when the test script is called. The datasets will be stored in data/ directory.

Demo

We provide a demo model in the models directory, which can be used to evaluate the results for CPD task.

python run_test.py ./models/cpd_model_sample.pt cpd

Acknowledgements

The following packages/libraries are adapted/communicated with in the codebase of GBPNet:

• Pytorch
• Pytorch Lightning
• Pytorch Geometric
• NumPy
• Catalyst
• Ingraham, et al, NeurIPS 2019
• Townshend et al, NeurIPS 2020
• Jing, et al, ICLR 2021

We thank their authors for providing the codebase.