LigBind

LigBind is a relation-aware framework with graph-level pre-training to enhance the ligand-specific binding residue predictions for 1159 ligands, which can effectively cover the ligands with a few known binding proteins.

We also release a ligand-general method LigBind-G for query ligand or general ligand binding residue prediction.

Preparation

LigBind is built on Python3. We recommend to use a virtual environment for the installation of GraphBind and its dependencies. A virtual environment can be created and (de)activated as follows by using conda:

# create
$ conda create -n LigBind_env python=3.6
# activate
$ source activate GraphBind_env

When you want to quit the virtual environment, just:

$ source deactivate

Download the source code of LigBind from GitHub:

$ git clone https://github.com/yingx/LigBind.git

Download all of the trained models from url.

$ tar zxvf LigBind_pth.tar.gz
$ cp -r LigBind_pth LigBind/checkpoints

Instead, you can download models for specific ligand in url. Taking ZN as an example:

$ mkdir LigBind/checkpoints/LigBind_pth
$ tar zxvf ZN.tar.gz
$ cp -r ZN LigBind/checkpoints/LigBind_pth

Install the dependencies as following:

$ pip install biopython
$ pip install torch==1.4.0
$ pip install torch-scatter==2.0.3
$ pip install torch-cluster==1.5.2
$ pip install torch-sparse==0.5.1
$ pip install torch-spline-conv==1.2.0
$ pip install torch-geometric==1.7.0
$ conda install -c conda-forge rdkit   # version 2020.09.1
$ pip install git+https://github.com/bp-kelley/descriptastorus

Install the bioinformatics tools:

(1) Install DSSP (version: 2.0.4) for extracting SS (Secondary structure) profiles

$ cd LigBind
$ chmod +x ./dssp

(2) Install HHblits for extracting HMM profiles

To install HHblits (version: 3.3.0) and download uniclust30_2018_08 for HHblits, please refer to hh-suite. Set the absolute paths of HHblits and uniclust30_2018_08 databases in the script "./prediction.py".

Usage

Method 1. LigBind

Prediction for ligands in the ligand-specific datasets with fine-tuned models.

If the target ligand is included in 1159 ligands, you can select the ligand type and use ligand-specific LigBind for prediction:

You can search ligand ID by ligand name from Ligand information.

If you want to make prediction for mmCIF files instead of PDB files, please install Open Babel and convert mmCIF into PDB with:

$ obabel output/example/7bv2.cif -icif -opdb -O output/example/7bv2.pdb

(Option) Since the most time-consuming step of prediction pipeline is generating MSAs with HHblits and computing the secondary structures with DSSP, users can generate them before prediction:

$ python extract_protein_features.py --querypath output/example  --protein_filename 7bv2.pdb --chainid A --method LigBind --ligands ADP,ATP --cpu 20

Make prediction:

$ python prediction.py --querypath output/example  --protein_filename 7bv2.pdb --chainid A --method LigBind --ligands ADP,ATP --cpu 20

Results are saved in output/example/results. The predicted probabilities and binary results are saved in csv file. We provide an annotated pdb for per query ligand with b-factor replaced by an indicator of results, so that the results could be easily displayed in molecular viewers such as pymol.

Method 2. LigBind-G

If the target ligand isn't included in 1159 ligands, you can input the ligand SMILES and use ligand-general LigBind-G for prediction:

$ python prediction.py --querypath output/example  --protein_filename 7bv2.pdb --chainid A --method LigBind-G --ligand_filename ligand_smiles.txt --cpu 20

Method 3. LigBind-G without ligand

If you want to predict general ligand-binding residues without ligand information, you can choose this method:

$ python prediction.py --querypath output/example  --protein_filename 7bv2.pdb --chainid A --method LigBind-G-nolig --cpu 20

Option: Construct your own dataset with low sequence identity.

First, MMseqs2 is applied for removing proteins of dataset1 with over 30% sequence identity to any proteins of dataset2:

$ ./mmseqs easy-search dataset1 dataset2 out tmp --min-seq-id 0.3

Then, CD-HIT (version: cd-hit-v4.8.1-2019-0228) is applied for reducing sequence identity of dataset1 itself. Following the CD-HIT user's guide, we reduce sequence identity of dataset1 to 30% with three steps:

$ ./cd-hit -i dataset1 -o dataset1_80 -c 0.8 -n 5 -d 0 -M 16000 -T 16
$ ./cd-hit -i dataset1_80 -o dataset1_60 -c 0.6 -n 4 -d 0 -M 16000 -T 16
$ ./cd-hit -i dataset1_60 -o dataset1_30 -c 0.3

License

All materials are made available under the terms of the Creative Commons Attribution-NonCommercial 4.0 International Public License (CC BY-NC 4.0) license. You can find details at: CCBY4.0.

Online service

Online retrieval service and benchmark datasets are in here.