April 8, 2005

This is material that concerns a comparison of the SAM and HMMER HMM
packages carried out by Markus Wistrand and Erik Sonnhammer. 
Please also see the paper: "Improved profile HMM performance by assessment of critical features in SAM and HMMER", Wistrand et al, BMC Bioinformatics, 6:99

Included material
- TEST.tar.gz: 
The testset that was used consists of 505 Pfam alignments. These were
matched to the superfamily level in SCOP. HMMs were trained from the
Pfam alignments and then tested on sequences from the ASTRAL set that
belong to the same superfamily (positives) or to another fold
(negatives). Zipped and tarred.

- INDEX:
The file specifies which positive and negative sequences that
are meant to use to test each HMM built from a particular alignment. 
> Pfam_fam_alignment  Test_seq_prefix average_length_of_seqs
average_identity_of_seqs
positive seq 1
positive seq 2
....

- HiddenMarkovModel_mw.pm and model_convert.pl:
This perl code that was used to convert between HMMER and SAM models. It is
almost identical to M. Madera and J. Gough's free code, but I have
included the possibility to convert to and from HMMER global/local
models. There is one or two additional minor changes (see the code) 
See Madera and Gough's excellent description!
http://www.mrc-lmb.cam.ac.uk/genomes/julian/convert/convert.html

- CODE.tar:
This is the my modification to the free HMMER code (2.3.2). It
includes an implementation of an entropy-based way of calculating the
effective sequence number. It also includes the atp algorithm which
trains the HMM on negative data. Download the file and do 'tar -xf'. 
Then see further in the README file in SAM_HMMER_code.

- recode3.20comp.pri:
This is a Dirichlet mixture prior estimated by the Computational
Biology group at UCSC. I have modified the format such that it can be
used in HMMER (--prior recode3.20comp.pri). It was used during the
test.

- recode3.20comp.fssp-trans.pri:
Here I have combined two priors from UCSC (the recode3.20comp.pri and
the fssp-trained.regularizer) to a priorfile that can be read by
HMMER. It was also used during the test.

OBS: The prior research at UCSC is free and  can be reached here:
http://www.cse.ucsc.edu/research/compbio/dirichlets/index.html


HMMER is distributed under the terms of the GNU General Public License as
published by the Free Software Foundation (see LICENSE distributed along with HMMER). Any modifications 
to HMMER must also be distributed under the same terms. This means that
this code is free to anyone and that any changes must also be made freely available.

See INSTALL for installation of SAM_HMMER_code (first you need to install HMMER 2.3.2 from 
http://hmmer.wustl.edu/).