DeepTMInter

tags: transmembrane protein predicting interaction sites v1.0

:information_source: News: The DeepTMInter Docker software has become available at deeptminter/releases. Please see below.

Overview

This repository is a software package of DeepTMInter. DeepTMInter is a deep-learning-based approach and it was developed using stacked generalization ensembles of ultradeep residual neural networks. The approach shows a substantial improvement for predicting interaction sites in transmembrane proteins compared to existing methods. All training and benchmarked data are available here and other data are made available upon requests of users via email.

System Requirement

We tested our software on a Linux operation system due to a number of Linux-dependent software packages generating input features. If you have the feature files as shown in ./input/, you are able to run our program on multiple platforms, e.g. Windows and Mac. Please be sure of python (version>3.5) installed before using. We highly recommend Anaconda, an integrated development environment of python, which eases the use and management of python packages.

Installation

1. install dependencies.

   • HHblits - generating multiple sequence alignments
   • Gaussian DCA - predictor of residue contacts
   • Freecontact - predictor of residue contacts
   • Phobius - predictor of transmembrane topologies

2. install protein sequence database.

   • Uniclust30 database - a curated protein sequence database based on UniProt for HHblits

3. install DeepTMInter

   • To download the prediction models here. Please put the models in folder deeptminter/model/.

   • To download a stable version of DeepTMInter here.

   • To obtain the latest version of DeepTMInter do

     git clone https://github.com/2003100127/deeptminter.git
     

4. install DeepTMInter of a Docker version (optional)

   • To install Docker here.

   • To download the five partitioned Docker packages here.

   • To use 7z to decompress the 5 partitioned Docker packages. This step will result in a file named deeptminter_10.docker.

   • to import deeptminter_10.docker by

     docker load < deeptminter_10.docker
     

   • to use deeptminter_10.docker by

     docker exec -it deeptminter_10.docker bash
     

5. install python dependencies

   pip install -r requirements.txt
   

Usage

1. src/troll.sh

   • description troll.sh is used to generate multiple sequence alignments, transmembrane topologies, and all of evolutionary coupling features including EVfold (generated using FreeContact) and Gaussian DCA.

   • shell commands

     • general (please specify the installed location of the executables or the database in Installations 1 and 2 and put your fasta sequence in the input path before running the following command.)

       ./troll.sh -n NAME -c CHAIN -i /YOUR/INPUT/PATH/
       

     • example

       ./troll.sh -n 3jcu -c H -i ./input/
       

   • parameters

     • required

       -n --name -> a sequence name.
       -c --chain -> a chain name
       -i --input -> input path
       

2. src/gdca.julia

   • description

     gdca.julia is used to generate Gaussian DCA file. You'd better run it cf. https://github.com/carlobaldassi/GaussDCA.jl.

3. run_deeptminter.py

   • description

     If you have the feature files shown in the ./input/ directory, you can skip over steps 1-2 to step 3. We tested this step in a rigorous way. Be sure of every feature file already in the ./input/ or your preferred input file path. Finally, it works easily like this.

   • python commands

     • general

       python run_deeptminter.py -n NAME -c CHAIN -i /YOUR/INPUT/PATH/ -o /YOUR/OUTPUT/PATH/ -r REGION
       

     • example

       python run_deeptminter.py -n 3jcu -c H -i ./input/ -o ./output/ -r transmembrane
       

   • parameters

     • required

       -n --name -> a sequence name. For example, '3jcu'.
       -c --chain -> a chain name. For example, 'H'. This can be empty if you prefer a sequnce name like '3jcuH' or '0868'.
       -i --input -> input path
       -o --output --> prediction results
       -r, --region --> region of transmembrane protein. It can take 'transmembrane', 'cytoplasmic', 'extracellular', 'combined', 'all where 'combined' means accumulation of 'transmembrane', 'cytoplasmic', 'extracellular'. 'all' means the whole fasta sequence.
       

4. description of output file

   It finally returns an output file with the suffix of .deeptminter.

   • The predictions of interaction sites in tansmembrane proteins are shown in the output file, with three columns: 1). positions of animo acids in the input sequence; 2) animo acids; 3) probabilities of being interaction sites.
   • Please note that if you have a sequence sharing a high sequence identity to the proteins in the TrainData dataset, we recommend that any of the three output files with the suffix '.mexpand1;.mexpand2;.mexpand3' would be the best option for you.
   • If you want to get the results in the context of no ideally preferred regions predicted by Phobius. You can set -r as combined to run the program. This will return the predictions of the whole fasta sequence. Then, you can tailor the whole predictions to whatever you want.

How to cite

J. Sun. D. Frishman. Improved sequence-based prediction of interaction sites in α-helical transmembrane proteins by deep learning. Comput. Struct. Biotechnol. J., 19:1512-1530, 2021. DOI: 10.1016/j.csbj.2021.03.005.

or

@article{DeepTMInter2021,
    title = {Improved sequence-based prediction of interaction sites in α-helical transmembrane proteins by deep learning},
    author = {Jianfeng Sun and Dmitrij Frishman},
    journal = {Computational and Structural Biotechnology Journal},
    volume = {19},
    pages = {1512-1530},
    year = {2021},
    issn = {2001-0370},
    doi = {https://doi.org/10.1016/j.csbj.2021.03.005},
    url = {https://www.sciencedirect.com/science/article/pii/S2001037021000775},
}

Contact

If you have any question, please contact Jianfeng Sun. We highly recommend creating issue pages when you have problems. Your issues will subsequently be responded.