DeepGlycanSite: A Highly Accurate Carbohydrate-binding Site Prediction Algorithm

Official implementation of DeepGlycanSite, a state-of-the-art method for carbohydrate-binding site prediction. This repository contains all code, instructions, and model weights necessary to run the method or to retrain a model. If you have any questions, feel free to open an issue or reach out to us: xicheng@simm.ac.cn

Description

DeepGlycanSite is an open-source method for carbohydrate-binding site detection, with or without known glycan. It can perform a whole range of carbohydrate-binding site prediction tasks.

Things DeepGlycanSite can do

• Discover common glycan binding site based on protein structure
• Discover binding site for specified glycan based on protein structure
• Guide mutation design for known glycan targets

---

Table of contents

1. Dataset
2. Setup Environment
   1. For CPU
   2. For GPU
3. Running DeepGlycanSite on test system
4. Retraining DeepGlycanSite
5. License

Dataset

If you want to train one of our models with the data then:

1. download pdb ids from dataset/
2. divide them into ligands and proteins
3. unzip the directory and place it into data such that you have the path data/PDBBind_processed

Setup Environment

We will set up the environment using Anaconda. Clone the current repo

git clone https://github.com/xichengeva/DeepGlycanSite.git

To use conda or mamba to create the environment, you can use:

cd DeepGlycanSite
conda env create -f environment.yml

This is an example of how to set up a working conda environment from scratch to run the code (but make sure to use the correct pytorch, pytorch-geometric, cuda versions, or cpu only versions):

conda create --name DeepGlycanSite python=3.9
conda activate DeepGlycanSite

For CPU:

pip install torch==1.13.1+cpu torchvision==0.14.1+cpu torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cpu
pip install torch_geometric==2.3.0
pip install "fair-esm[esmfold]"
pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-1.13.1+cpu.html
pip install pytorch-lightning==1.9.3 &&    pip install h5py==3.8.0 && pip install rdkit==2022.9.5 && pip install Bio==1.5.5 && pip install pandas==1.5.3 &&  pip install MDAnalysis==2.4.2  && pip install pymatgen==2023.7.20  && pip install tokenizers==0.13.3 && pip install lmdb==1.4.1 && pip install addict==2.4.0 && pip install transformers==4.30.0
pip uninstall -y numpy 
pip install numpy==1.26.1

For GPU:

pip install torch==1.13.1+cu116 torchvision==0.14.1+cu116 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu116
pip install torch-geometric==2.3.0
pip install "fair-esm[esmfold]"
pip install torch-scatter torch-sparse torch-cluster torch-spline-conv torch-geometric -f https://data.pyg.org/whl/torch-1.13.1+cu116.html
pip install pytorch-lightning==1.9.3 && pip install h5py && pip install rdkit==2022.9.5 && pip install Bio==1.5.5 && pip install pandas==1.5.3 &&  pip install MDAnalysis==2.4.2  && pip install pymatgen==2023.7.20  && pip install tokenizers==0.13.3 && pip install lmdb==1.4.1 && pip install addict==2.4.0 && pip install transformers==4.30.0
pip uninstall -y numpy 
pip install numpy==1.26.1

Installation may take 1 hours on a normal desktop computer.

In case any version is incompatible, check the environment.yml file.

Running DeepGlycanSite on test system

The protein inputs need to be .pdb files. The ligand input needs to be .sdf files. Both the files need HYDROGENS to be added.

Run inference with only receptor:

python single_case_prediction.py --conf P2Y14_example/hparams_rec.yaml  --ckpt_path ckpts/rec_only.ckpt --input_fn P2Y14_example/P2Y14_AF.pdb  --out_path P2Y14_example/ --output_fn P2Y14_af.txt

P2Y14_example/P2Y14_af.txt will be the output file and the second column is the probability for each residue. The closer to 1, the higher the likelyhood of interacting with carbohydrates. The Nth line means the Nth residue in the protein.

Run inference with receptor and ligand:

python single_case_prediction.py --conf P2Y14_example/hparams.yaml  --ckpt_path ckpts/with_ligand.ckpt --input_fn P2Y14_example/P2Y14_AF.pdb,P2Y14_example/GDP.sdf --out_path P2Y14_example/ --output_fn P2Y14_af_GDP.txt

Use a comma to connect pdb and ligand sdf to activate DeepGlycanSite+ligand model, P2Y14_example/P2Y14_af_GDP.txt will be the output file

If auto-download fails or you meet problem with "no checkpoint file", you can also download the checkpoints from https://huggingface.co/Xinheng/DeepGlycanSite/tree/main or https://zenodo.org/records/10065607, just put rec_only.ckpt and with_ligand.ckpt to ./ckpt, and mol_pre_all_h_220816.pt to ./src/unimol_tools/weights

Running per line takes 1-3 minutes on a normal desktop computer.

Retraining DeepGlycanSite

Use single_case_prediction.py for per protein and ligands, then combine the .pt files in out_path, and you can train it with DeepGlycanSite.py or DeepGlycanSite_lig.py with following codes:

python DeepGlycanSite.py --conf P2Y14_example/hparams_rec.yaml --output-path DeepGlycanSite_rec

python DeepGlycanSite_lig.py --conf P2Y14_example/hparams.yaml --output-path DeepGlycanSite_lig

License

Apache License 2.0