TMP-SurResD

A deep-learning framework, TMP-SurResD (Transmembrane Proteins Surface Residue Distance Prediction), for simultaneously predicting relative distance of functional surface residues based on the combination of co-evolutionary information.

The workflow and architecture of TMP-SurResD

Details of the TMP-SurResD framework

Download data

We provide all the transmembrane protein sequences used in the manuscript, which are available in the './fasta/' directory.

Quick Start

Here we provide the three trained models described in the manuscript, namely "24_ccmpred_90", "24_hhm_ccmpred_90", and "24_onehot_hhm_ccmpred90". The input dimension of model "24_ccmpred_90" is L×L×1 (CCM feature), "24_hhm_ccmpred_90" is L×L×61 (onehot+HHM), and "24_onehot_hhm_ccmpred90" is L×L×101 (onehot+HHM+CCM). The test data provided by us can be used to verify our proposed method.

Requirements

Tools used in this study can be publicly available online:
Pytorch (https://pytorch.org/);
Python 3.6 (https://www.python.org/);
PDBTM (http://pdbtm.enzim.hu);
Biopython (https://biopython.org/);
CD-HIT (http://weizhong-lab.ucsd.edu/cd-hits/);
TMP-SSurface-2.0 (https://github.com/NENUBioCompute/TMP-SSurface-2.0);
HHblits (http://toolkit.genzentrum.lmu.de/hhblits/);
DeepMSA (https://seq2fun.dcmb.med.umich.edu/DeepMSA/);
CCMpred (https://bitbucket.org/soedinglab/ccmpred);
PSI-BLAST (ftp://ftp.ncbi.nlm.nih.gov/blast/executables/blast+/LATEST/).

Test & Evaluate in Command Line

We provide test.py that is able to run pre-trained models. Run it with:

python test.py 

It is important to note that the 'test.py' provided is only to verify experimental results recorded in the manuscript and cannot be used directly to predict unknown protein sequences. If you need to predict an unknown protein sequence, the input data needs to be prepared in advance. Refer to the data feature extraction process shown below:

As shown in the figure, TMP-SSurface2 for generating rASA; HHblits for generating HHM features; deepMSA for generating MSAs; CCmpred for generating CCM.

1. If you want to use the "24_ccmpred_90" model, just prepare the CCM features run out with CCMpred in advance. Then select three lines of code in the test.py file.

x_test, y_test = con.main('ccmpred')
model_path = './model/24_ccmpred_90'
n = 1

2. If you want to use the "24_hhm_ccmpred_90" model, just prepare the CCM features run out with CCMpred and the HHM features generated with HHBlits in advance. Then select three lines of code in the test.py file.

x_test, y_test = con.main('ccmpred')
model_path = './model/24_hhm_ccmpred_90'
n = 61

3. If you want to use the "24_onehot_hhm_ccmpred90" model, just prepare the CCM features run out of CCMpred, the HHM features generated by HHBlits, and the onehot encoding in advance. Then select three lines of code in the test.py file.

x_test, y_test = con.main('ccmpred')
model_path = './model/24_onehot_hhm_ccmpred90'
n = 101

Progress

README for running TMP-SurResD.