Requirements
------------
The Jar-file used for predicting intramolecular contacts from amino acid sequence needs Java 
to be downloaded and installed to work.

The Jar-file also needs the files in the /final_HMM/ directory to work, these are included in 
the FragHMMent.tar.gz file.

To unzipp, type: tar xfz FragHMMent.tar.gz

Input data
-----------
The Jar-file uses a profiles as input data (targetname.mtx), these can be generated using 
psi-blast and makemat (http://www.ncbi.nlm.nih.gov/BLAST/download.shtml).

The method also needs predicted secondary structure to predict contacts 
(e.g. PSIPRED: http://bioinf.cs.ucl.ac.uk/psipred/). The data should be in the format seen 
below and has to be a textfile (ss_targetname.txt).

>targetname
SMKPKLCRLAKGENGYGFHLNAIRGLPGSFIKEVQKGGPADLAGLEDEDVIIEVNGVNVLDEPYEKVVDRIQSSGKNVTLLVCGKKAQDTV
CCCCEEEEEEECCCCCCEEEEECCCCCCEEEEEECCCCHHHHCCCCCCCEEEEECCEECCCCCHHHHHHHHHHCCCEEEEEEECCCCCCCC

The Jar-file will only work if there are two target files in the same directory, one should 
be named targetname.mtx (the profile) and the other should be named ss_targetname.txt (the 
file containing the predicted secondary structure).

Examples of profile and secondary structure files can be found in the /test/ directory.

NOTE: The directory /perl/ contains a perl-script that runs the Jar-file together with PSI-BLAST
and PSIPRED given that these programs are installed. See the README file for details.

Usage
-----
to run the Jar-file type

	java -jar FragHMMent.jar targetname

in the command console.

The program will produce a file named targetname.txt (see the /test/ directory for an 
example) with residue-residue contact predictions in the CASP format: 
see http://www.predictioncenter.org/casp7/doc/casp7-format.html#RR.

Tips
----
For best possible performance use the Non-redundant database for generating profiles and 
secondary structure.

Citing
------
If you use this method in your work, please cite: P. Björkholm, P. Daniluk, A.  Kryshtafovych, 
K. Fidelis, R. Andersson  and T. R. Hvidsten. Using multi-data hidden Markov models trained 
on local  neighborhoods of protein structure to predict residue-residue contacts. Currently 
awaiting publication.

Problems
--------
If you are having problems running this jar-file please contact pbh@sbc.su.se.