Click here to download DBNN source codes.

Click here to download datasets used to evaluate DBNN.

See below for the help of installation and usage.

 

 

 

 

INSTALLATION

=============

 

->Environment

 

1. Operating system:

  *   LINUX / UNIX

 

2. Requisite softwares:

  *   MATLAB (v6.5 or later)

  *   Bayes Net Toolbox (BNT);

      can be downloaded from http://bnt.sourceforge.net/

 

3. Optional software:

  *  NCBI BLAST suite;

     can be downloaded from ftp://ftp.ncbi.nih.gov/toolbox/

 

NOTE:

  If you have not installed BLAST locally, the way to use DBNN is:

  running BLAST and constructing your own PSSM file somewhere else

  (e.g.  online BLAST), and then provide the PSSM file to DBNN.

 

    

->Compilation

 

Type the following commands in the home directory of DBNN to compile

the package:

 

  make

  vi rundbnn

  (modify the first FOUR variables in the rundbnn file, i.e. homedir,

  matlabdir, blastdir, and dbname, according to your own

  configurations and exit)

  ./rundbnn

  (the first run to check the installation of requisite softwares)

 

NOTE:

  If you want to launch DBNN in directories other than the home

  directory of DBNN, you need to add the home directory of DBNN into

  the search path of MATLAB.

 

 

 

USAGE

=======

 

->Quick start

 

The simplest way to use DBNN is typing:

 

   ./rundbnn seq.fasta output

 

where seq.fasta is your amino acid sequences saved in FASTA format,

and output is the prefix of prediction files (typically, two

prediction files will be generated:

 

      output.fasta

 

and

 

      output.raw

 

the former contains the predictions in FASTA format, and the latter

contains scores of secondary structures for each residue site). An

additional file generated by launching rundbnn

is

 

      query.pssm

 

which is the PSSM file corresponding to your seq.fasta.

 

 

->Run with non-default parameters

 

By default, rundbnn reads parameters from seven files associated

with DBNN package:

 

   default.M1.mat

   default.M2.mat

   default.M3.mat

   default.M4.mat

   default.nn1.lin

   default.nn1.sig

   default.jury

 

You can re-train the DBNN (see below) and provide your own parameter

files when launching rundbnn. For example, your parameter files are

named with prefix "newpara" (i.e. newpara.M1.mat, ... newpara.nn1.lin,

... newpara.jury), you can give them to rundbnn by typing:

 

  ./rundbnn seq.fasta output newpara

 

If you have not installed BLAST, and instead you have run BLAST and

constructed the PSSM file somewhere else (e.g. online), you can use

DBNN by typing:

 

   ./rundbnn yourPSSM output [ prefix-of-parameter-files ]

 

 

->Run DBN and NN separately

 

1. Run DBN

type:

  

   ./rundbnn -dbn seq.fasta output [ prefix-of-parameter-files ]

 

2. Run NN

type:

 

  ./rundbnn -nn seq.fasta output [ prefix-of-parameter-files ]

 

 

->Training of DBN, NN, and jury-NN

 

1. Training of DBN

type:

 

  ./dbntrain yourPSSM secstr.fasta parameters

 

where secstr.fasta is the annotation of secondary structure for your

proteins (in FASTA format). Here you must always provides the PSSM

file and the prefix of the parameter files: "paramters".  Four files

will be generated:

 

   parameters.M1.mat

   parameters.M2.mat

   parameters.M3.mat

   parameters.M4.mat

 

 

2. Training of NN

type:

 

   ./nntrain yourPSSM secstr.fasta parameters

 

Two files will be generated:

  

   parameters.nn1.lin

   parameters.nn1.sig

 

3. Training of jury-NN

type:

 

   ./jurytrain yourPSSM secstr.fasta parameters

 

The file generated is:

 

   parameters.jury

 

 

=================

Xin-Qiu Yao

Jan. 7, 2007