GraphBepi is a novel graph-based method for accurate B-cell epitope prediction, which is able to capture spatial information using the predicted protein structures through the edge-enhanced deep graph neural network.
We recommend you to use the web server of GraphBepi if your input is small.
GraphBepi is developed under Linux environment with:
- python 3.9.12
- numpy 1.21.5
- pandas 1.4.2
- fair-esm 2.0.0
- torch 1.12.1
- pytorch-lightning 1.6.4
- (optional) esmfold
To run the full & accurate version of GraphBepi, you need to make sure the following software is in the mkdssp directory:
DSSP (dssp ver 2.0.4 is Already in this repository)
git clone https://github.com/biomed-AI/GraphBepi.git && cd GraphBepipython dataset.py --gpu 0
It will take about 20 minutes to download the pretrained ESM-2 model and an hour to build our dataset with CUDA.
After building our dataset BCE_633, train the model with default hyper params:
python train.py --dataset BCE_633
- Please execute the following command directly if you can provide the PDB file.
- If you do not have a PDB file, you can use AlphaFold2 to predict the protein structure.
python test.py -i pdb_file -p --gpu 0 -o ./output
or
We have also deployed a faster structural prediction model ESMFold in our project, so you can process the sequences directly by following the commands below.
python test.py -i fasta_file -f --gpu 0 -o ./output
The GrpahBepi web server is freely available: interface
Citation:
@article{zengys,
title={Identifying the B-cell epitopes using AlphaFold2 predicted structures and pretrained language model},
author={Yuansong Zeng, Zhuoyi Wei, Qianmu Yuan, Sheng Chen, Weijiang Yu, Jianzhao Gao, and Yuedong Yang},
journal={biorxiv},
year={2022}
publisher={Cold Spring Harbor Laboratory}
}
Contact:
Zhuoyi Wei ([email protected])
Yuansong Zeng ([email protected])