Jnet: A Neural Network Protein Secondary Structure Prediction Method

Quickstart (!)

Installation:

  1. Firstly set an environment variable JNET to point to the installation directory,
i.e. if installed in /homes/james/
     setenv JNET /homes/james/jnet (for sh, csh, tcsh)
   or
     export JNET=/homes/james/jnet (for bash)
  2. All the perl scripts in the perl dir will need there #!CHANGEJNET headers changing to direct the script to the perl interuptor, which is usually /usr/loca/bin/perl. The script $JNET/perl/change_header will do this for you if you have sed and sh. Change directory to the Perl installation directory and run the script from there. Perl 5 is a requirement for all the scripts in Jnet.
  3. *SGI ONLY* If using an older SGI machine (r4000), rename the jnet_mips2 binary to jnet. The original jnet binary was optimised for the R10,000, mips4 architecture.

Execution:

  1. Jnet needs 4 files:
    1. A FASTA format multiple sequence alignment
    2. A HMMER2 profile
    3. A PSIBLAST PSSM profile
    4. A PSIBLAST frequency profile
    either a, a+b, or a+b+c+d can be applied.
  2. To carry out a prediction you will need to do the following:
    1. Take your target sequence (my.seq)
    2. Search sequence against database with PSIBLAST
      blastpgp -j3 -d swissprot -i my.seq > my.blast (psiblast 2.0.3)
      blastpgp -Q my.pos -m6 -b20000 -j3 -d swissprot -i my.seq > my.blast (psiblast 2.0.10)

    3. Fetch the sequences from the BLAST report with your favorite method. Call them my.seqs.
    4. Build a multiple sequence alignment from my.seqs with CLUSTALW (v1.7 or greater):
      clustalw -infile=my.seqs -outfile=my.msf -output=gcg -outorder=input
    5. If you have version 2.0.3 of PSIBLAST run the getfreq and getpssm perl scripts that come with the Jnet distribution on the BLAST report file
      getfreq my.blast
      getpssm my.blast
    6. If you have version 2.0.10 of PSIBLAST, thing are slightly different. You'll need the updated parsers from here. This extra step is due to changes in the PSIBLAST output. Once you have these then run:
      parse_psi -freq my.blast > my.blast.freq
      getpssm my.pos > my.blast.pssm
    7. Run: msf2jnet my.msf (this generates the FASTA file for Jnet)
      gethmm my.msf (this generates a HMM based profile for Jnet)
    8. Run Jnet:
      jnet -p my.msf.fa (for prediction from the alignment only)
      jnet -p my.msf.fa my.msf.hmmprof (for prediction without the PSIBLAST profiles)
      jnet -p my.fa my.msf.hmmprofmy.blast.pssm my.blast.freq (if you have the PSIBLAST profiles)
    [Back]
    www-compbio@compbio.dundee.ac.uk