Introduction

DeepProSite is a framework for identifying protein binding site that utilizes protein structure and sequence information. DeepProSite is fast and accurate, and it's easy to install and run. The DeepProSite web server is freely available in here.

System requirement

python 3.7.11
numpy 1.21.2
pandas 1.1.3
torch 1.8.0
biopython 1.79
sentencepiece 0.1.96
transformers 4.15.0

Install and set up DeepProSite

1. Clone this repository by git clonehttps://github.com/WeiLab-Biology/DeepProSite.git or download the code in ZIP archive. The latest version is DeepProSite-main.

2. Download the ESMFold model and install according to the official tutorials：

pip install "fair-esm[esmfold]"
pip install 'dllogger @ git+https://github.com/NVIDIA/dllogger.git'
pip install 'openfold @ git+https://github.com/aqlaboratory/openfold.git@4b41059694619831a7db195b7e0988fc4ff3a307'

Predict structures via python scripts/fold.py -i ./datastes/test_protein.fa -o ./datasets/pdb --cpu-offload

3. Download the pre-trained ProtT5-XL-UniRef50 model in here. The downloaded model is stored in ./pretrained_model/Rostlab/prot_t5_xl_uniref50. Extracting language model features and normalize them based on the training set：

python ./ProtTrans.py --fasta ./datasets/test_protein.fa --out_path ./feature/rawembedd --gpu 0
python ./process_ProtTrans.py

4. Add permission to execute for DSSP and extract dssp:

chmod +x ./Software/dssp-2.0.4/mkdssp
python ./get_dssp.py 

5. Merge features:

python ./pad_feature.py 

Run DeepProSite for prediction

Run the following command to predict the binding sites of the sequence in "./datasets/example.fa" on GPU (id=0):

CUDA_VISIBLE_DEVICES=0 python ./main.py --task PRO --test --seed 2022 --run_id prediction --dataset_path ./datasets/ --feature_path ./feature/

Datasets and models

The datasets used in this study are stored in ./datasets/
The trained DeepProSite models can be found in ./model/

Citation and contact

@article{Fang2023DeepProSiteSP,
  title={DeepProSite: Structure-aware Protein Binding Site Prediction Using ESMFold and Pretrained Language Model.},
  author={Yitian Fang and Yi Jiang and Leyi Wei and Qin Ma and Zhixiang Ren and Qianmu Yuan and Dong-Qing Wei},
  journal={Bioinformatics},
  year={2023},
  url={https://api.semanticscholar.org/CorpusID:265497196}
}