ProB-site

Intro

ProB-site is a CNN model that predicts binding sites of protien protein interactions. It utilizes evolutionalry information and predicted secondary struacture information extracted from protein sequences.

System Requirment

This model has been developed in Linux environment with:

• python 3.8.10
• numpy 1.21.5
• pandas 1.4.2
• tensor flow 2.3.0

Software and Database requirments

To run full version of the ProB-site, it requires following software to extract features

• Blast+ and UniRef90
• HH-suite and Uniclust30
• DSSP

Built software and Database

• Install DSSP
  install libraries

  sudo apt-get install libboost-all-dev  
  sudo apt-get install -y libz-dev  
  sudo apt-get install -y libbz2-dev  
  sudo apt-get install -y automake  
  sudo apt-get install -y autotools-dev  
  sudo apt-get install -y autoconf 
  

  From link DSSP download dssp-3.1.4.tar.gz
  unzip it using command
  tar -zxvf dssp-3.1.4.tar.gz
  compile the program using follwoing command

  cd dssp-3.1.4  
  ./autogen.sh  
  ./configure  
  make 
  

  Here dssp-3.1.4/mkdssp is DSSP software path

• Install Blast+ and database
  download [uniref90.fasta.gz](link: https://ftp.uniprot.org/pub/databases/uniprot/uniref/uniref90/)
  unzip downloaded niref90.fasta.gz file
  gzip -d uniref90.fasta.gz
  download [ncbi-blast-2.13.0+-x64-linux.tar.gz](link https://ftp.ncbi.nlm.nih.gov/blast/executables/blast+/LATEST/)
  unzip ncbi-blast-2.13.0+-x64-linux.tar.gz
  tar -zxvf ncbi-blast-2.13.0+-x64-linux.tar.gz
  In created folder run following command, making sure ./uniref90.fasta proper path is given
  ./ncbi-blast-2.13.0+/bin/makeblastdb -in ./uniref90.fasta -parse_seqids -blastdb-version 5 -title "unirefdb" -dbtype prot The output of makeblastbd command will give the PSIBlast Database
  Here ncbi-blast-2.10.1+/bin/psiblast is PSIBLAST software path

• Install HH-suite and database

  download [uniclust30_2017_10.tar.gz](link: http://gwdu111.gwdg.de/~compbiol/uniclust/2017_10/) and [uniclust30_2017_10_hhsuite.tar.gz](link: http://gwdu111.gwdg.de/~compbiol/uniclust/2017_10/) and
  [uniclust_uniprot_mapping.tsv.gz](link: http://gwdu111.gwdg.de/~compbiol/uniclust/2017_10/)

  unzip downloaded file

  tar -zxvf uniref30_2017_10.tar.gz
  tar -zxvf uniclust30_2017_10_hhsuite.tar.gz
  tar -zxvf uniclust_uniprot_mapping.tsv.gz
  

  The output of above commands will give the HH-Suite Database
  Pre-requisits, install following dependencies

  sudo apt install pigz
  sudo apt install libopenmpi-dev
  sudo apt install sed
  sudo apt install md5deep
  sudo apt install clustalo
  sudo apt install kalign
  sudo apt install gawk
  sudo apt install node-connect-timeout  
  sudo apt-get install tar 
  

  install HH-suite software using follwing commands

  git clone https://github.com/soedinglab/hh-suite.git
  mkdir -p hh-suite/build && cd hh-suite/build
  cmake -DCMAKE_INSTALL_PREFIX=. ..
  make -j 4 && make install
  export PATH="$(pwd)/bin:$(pwd)/scripts:$PATH"
  

  Here hhsuite-3.0.3/build/bin/hhblits is HHsutie software path

• Updating software and data bases path in software
  In software open featuer_extarction.py and update paths according to your device

  In line 33 give correct path to data_path='./data_ext/'  
  In line 35 give correct path to dssp = Software_path + "dssp-3.1.4/mkdssp"  
  In line 36 give correct path to PSIBLAST = Software_path + "ncbi-blast-2.13.1+/bin/psiblast"  
  In line 37 give correct path to HHBLITS = Software_path + "hhsuite-3.0.3/build/bin/hhblits"  
  In line 38 give correct path to UR90 = "./unirefdb/uniref90.fasta"   
  In line 39 give correct path to HHDB = "./uniclust30_2017_10"
  

  In software open prediction.py and update paths according to your device

  In line 28 give correct path to  pre_path='./Feature/'  
  In line 29 give correct path to fea_path='./data_ext/'  
  

Run ProB-site for prediction

For prediction of binding site in a protein run following command:
python predictor.py -p 3zeu -c D
Here 3zeu is PDB_ID and D is chain of that PDB_ID, program will download necessary PDB file from online database

Data, feature and model

we have provided pre-computed feature and a pretrained model for those interested in reproducing the paper
List of Dataset used in this research in present in Featrues/data_seq folder
Features are stored in numpy format
Program can be run using pre-computed features without installing Blast+, HH-suite, and DSSP software. However binding sites of proteins present in pre-computed features list can only be predicted

Cite

For citation use the given BibTeX format
@article{khan2022prob,
title={ProB-Site: Protein Binding Site Prediction Using Local Features},
author={Khan, Sharzil Haris and Tayara, Hilal and Chong, Kil To},
journal={Cells},
volume={11},
number={13},
pages={2117},
year={2022},
publisher={MDPI}
}