MapPred

0. Requirements

• Tensorflow (1.6+)
  
• Keras (2.1.3+)
  
• Python (2.7+ or 3.5+)
  
• Numpy
  
• h5py
  
• PSIPRED
  
• solvpred
  
• CCMpred
  

1. Download the package

git clone https://github.com/AtlasWuqi/MapPred.git

2. Compile

NOTICE: This step can be skipped if OS is Ubuntu.
cd src
make
make install
cd ..

3. Get started

usage: ./bin/mappred.sh alnfile [outputfile] [cuda]
example: ./bin/mappred.sh ./example/1NH2C.aln

Input:

• xxx.ss2 : secondary structure predicted by PSIPRED
  
• xxx.solv : relative solvent accessibility predicted by solvpred
  
• xxx.ccmpred : contact map predicted by CCMpred
  
• xxx.aln : multiple sequence alignments(MSA)
  Format Requirement for an MSA:
  Please check out the example file (./example/1NH2C.aln) for a reference
  Each line in the MSA is the protein primary sequence. All sequences shall have the same length when gaps are also counted. The first sequence is the query sequence for which you would like to predict contacts. It shall not contain any gaps. The other sequences may contain gaps represented by '-'.
  NOTICE: make sure .ss2 .solv .ccmpred ready under the same directory of .aln
  

Output:

• xxx.mappred : final contact map predicted by DeepMeta [text format]
  
• xxxdeepmsa.npy : contact map predicted by DeepMSA [Numpy format]
  
• xxx.231stats : compressed covar matrices computed from the covariance matrices of the MSA
  
• xxx.frobstats : frobenius norm matrice computed from the covariance matrices of the MSA
  

a example can be found in file: ./start.sh
you can check the outputfile run by yourself and ./example/1NH2C.mappred.

Reference

Q Wu, Z Peng, I Anishchenko, Q Cong, D Baker, J Yang, Protein contact prediction using metagenome sequence data and residual neural networks, Bioinformatics, submitted (2018).