MEMSAT-SVM Usage Notes
======================

Program  and documentation  is  Copyright  (C)  2008  David T. Jones and
Timothy Nugent, all rights reserved.

All Trademarks and Registered Names are acknowledged in this document.

THIS SOFTWARE MAY ONLY BE USED FOR NON-COMMERCIAL PURPOSES. PLEASE CONTACT
THE AUTHOR IF YOU REQUIRE A LICENSE FOR COMMERCIAL USE.

Pre-Compilation
===============

You first need to get the large model files from our download site and
unpack them to a models/ directory.

From the root dir of your memsatsvm installation

wget http://bioinfadmin.cs.ucl.ac.uk/downloads/memsat-svm/models.tar.gz

tar -zxvf models.tar.gz

Compiling MEMSAT-SVM
====================

C and C++ compilers differ from system to system. However on a standard
Unix or Linux system, MEMSAT-SVM can be compiled simply with:

make


A copy of SVM Light is included; the executable will be placed in the bin
folder where the run_memsat-svm.pl expects to find it. Full details of
SVM light, including the licence, can be found at:

http://svmlight.joachims.org/


To produce graphical representations of topology, the GD library and the
perl GD module must be installed. The GD library should be present on
most modern Linux distributions. The perl GD module can usually be
installed by running the following command as root:

cpan -i GD


Configuring MEMSAT-SVM
======================

Importantly, a few paths need to be set before using MEMSAT-SVM: see the
comments on top of script "run_memsat-svm.pl" and set variables accordingly.

In particular, the absolute path to the Perl library needs to be set.

Also, the path to this directory, to the NCBI binary directory and to a
database for PSI-BLAST searches must be set.

my $mem_dir = '........'; # directory where "run_memsat-svm.pl" can be found
.
.
.
my $ncbidir = '........'; # where blastpgp and makemat are found
my $dbname  = '........'; # e.g. swissprot.fa, which has been formatdb'ed

You can also pass these last 3 values using the following paramaters:

./run_memsat-svm.pl -w /path/to/memsat-svm/ -n /usr/local/blast/bin/ -d swissprot.fa fasta.fa

If the NCBI directory is not specified, the script will try to find it using
'locate blastpgp'.

Finally, note that MEMSAT-SVM writes several temporary files (these will be
erased if flag '-e 1' is used), some of them inside the "input" folder, some
inside the "output" folder where the results are saved as well.
Default folders are present; however, when running several similar sequences
at the same time, it is wise to specify sequence-specific input and output
folders by means of flags '-i' and '-j', in order to avoid clashes due to
identical temporary file names.

Try "./run_memsat-svm.pl" or "./run_memsat-svm.pl -h" for help on all
available flags.


Ubuntu Configuration
====================

The ./run_memsat-svm.pl uses bash instead of Ubuntu's dash shell. You'll
have to change it like this:

sudo ln -sf /bin/bash /bin/sh


Running MEMSAT-SVM
==================

To run MEMSAT-SVM using fasta files:

./run_memsat-svm.pl examples/*fa


This will call PSI-BLAST in order to generate matrix files. If you have
already generated these, you can pass the -mtx flag to skip the PSI-BLAST
step:

./run_memsat-svm.pl -mtx 1 examples/*mtx


To perform a constrained prediction, Pass a constraints file as an argument.
This should have the same name as the fasta file but with a .constraints
suffix. You can still process multiple files at once; the constraints file
will be matched to the corresponding fasta or matrix file:

./run_memsat-svm.pl -mtx 1 examples/1R3J_C.constraints examples/1R3J_C.mtx


The constraints file should have the following format, where s,o,m,i
are signal peptide, outside loop, membrane and inside loop:

s:   1-15
o:   1-30
m:   37-59,82-100
i:   65,80,220-230


To run globmem-svm - in order to discriminate between globular and
transmembrane proteins - pass the -p flag:

./run_memsat-svm.pl -mtx 1 -p 1 examples/*mtx


Below is a full list of command line paramaters:

-p <0|1|2>     Programs to run. memsat-svm predicts topology, globmem-svm
               discriminates between transmembrane and globular proteins. Default 0.
               0 = Run memsat-svm
               1 = Run memsat-svm and globmem-svm
               2 = run globmem-svm
-mtx <1|0>     Process PSI-BLAST .mtx files instead of fasta files. Default 0.
-n <directory> NCBI binary directory (location of blastpgp and makemat)
-d <path>      Database for running PSI-BLAST.
-e <0|1>       Erase intermediate files. Default 0.
-g <0|1>       Draw topology schematic and cartoon. Default 1.
-m <int>       Minimum score for a transmembrane helix. Default: 220000
-r <int>       Minimum score for a re-entrant helix.    Default: 178000
-h <0|1>       Show help. Default 0.


Example Results
===============

In this  example MEMSAT-SVM  is used to predict the secondary  structure and
topology of Potassium channel KcsA.  The input file  (in FASTA format) is as
follows:

>1M57_H
MRHSTTLTGCATGAAGLLAATAAAAQQQSLEIIGRPQPGGTGFQPSASPVATQIHWLDGFILVIIAAITIFVTL
LILYAVWRFHEKRNKVPARFTHNSPLEIAWTIVPIVILVAIGAFSLPVLFNQQEIPEADVTVKVTGYQWYWGYE
YPDEEISFESYMIGSPATGGDNRMSPEVEQQLIEAGYSRDEFLLATDTAMVVPVNKTVVVQVTGADVIHSWTVP
AFGVKQDAVPGRLAQLWFRAEREGIFFGQCSELCGISHAYMPITVKVVSEEAYAAWLEQARGGTYELSSVLPAT
PAGVSVE

./run_memsat-svm.pl -p 1 examples/1M57_H.fa

The following output is produced by the run_memsat-svm.pl script (note the
paths to the NCBI directory and database will vary):


MEMSAT-SVM: Alpha-helical transmembrane protein topology prediction
using Support Vector Machines

Running PSI-BLAST: examples/1M57_H.fa
/usr/local/blast/bin/blastpgp -a 1 -j 2 -h 1e-3 -e 1e-3 -b 0 -d databases/uniprot_sprot -i memsat-svm_tmp.fasta -C
memsat-svm_tmp.chk >& memsat-svm_tmp.out

Generating SVM input files...
Running svm_classify...
bin/svm_classify -v 0 input/1M57_H_w33_GM.input models/MEMSAT-SVM_w33_GM.model output/1M57_H_SVM_w33_GM.prediction
bin/svm_classify -v 0 input/1M57_H_w35_TM.input models/MEMSAT-SVM_w35_IO.model output/1M57_H_SVM_w35_IO.prediction
bin/svm_classify -v 0 input/1M57_H_w33_TM.input models/MEMSAT-SVM_w33_HL.model output/1M57_H_SVM_w33_HL.prediction
bin/svm_classify -v 0 input/1M57_H_w27_TM.input models/MEMSAT-SVM_w27_SP.model output/1M57_H_SVM_w27_SP.prediction
bin/svm_classify -v 0 input/1M57_H_w27_TM.input models/MEMSAT-SVM_w27_RE.model output/1M57_H_SVM_w27_RE.prediction

Parsing SVM output files...
Running GLOBMEM-SVM...
Running MEMSAT-SVM...
bin/memsat-svm -m 220 -r 178 -f 1 output/1M57_H_SVM_ALL.out > output/1M57_H.memsat_svm

Written file output/1M57_H.memsat_svm
Written file output/1M57_H_schematic.png
Written file output/1M57_H_cartoon_memsat_svm_.png
Written file output/1M57_H.globmem_svm


DISCRIMINATING TRANSMEMBRANE PROTEINS FROM GLOBULAR
===================================================

The contents of 1M57_H.globmem_svm shows that the sequence is predicted to be a
transmembrane protein, with 49 residues predicted to lie within the membrane:

Transmembrane residues found:   45
Transmembrane score:            74.545051921
This looks like a transmembrane protein.


TOPOLOGY PREDICTION
===================

The contents of 1M57_H.memsat_svm is as follows:

...

Processing 1 helix:
Transmembrane helix 1 from  60 (in) to  81 (out) :	3903.24
Score = 3.772554

Processing 1 helix:
Transmembrane helix 1 from  60 (out) to  81 (in) :	3940.16
Score = 4.070846

...

Processing 5 helices:
Transmembrane helix 1 from  19 (in) to  34 (out) :	-445.657
Transmembrane helix 2 from  60 (out) to  81 (in) :	3940.16
Transmembrane helix 3 from  99 (in) to 119 (out) :	3253.82
Transmembrane helix 4 from 193 (out) to 208 (in) :	-506.732
Transmembrane helix 5 from 250 (in) to 265 (out) :	-565.749
Score = -100000.000000

Processing 4 helices:
Transmembrane helix 1 from  60 (out) to  81 (in) :	3940.16
Transmembrane helix 2 from  99 (in) to 119 (out) :	3253.82
Transmembrane helix 3 from 193 (out) to 208 (in) :	-506.732
Transmembrane helix 4 from 250 (in) to 265 (out) :	-565.749
Score = -100000.000000

Processing 3 helices:
Transmembrane helix 1 from  19 (in) to  34 (out) :	-445.657
Transmembrane helix 2 from  60 (out) to  81 (in) :	3940.16
Transmembrane helix 3 from  99 (in) to 119 (out) :	3253.82
Score = -100000.000000

Processing 2 helices:
Transmembrane helix 1 from  60 (out) to  81 (in) :	3940.16
Transmembrane helix 2 from  99 (in) to 119 (out) :	3253.82
Score = 7.337172

....

Processing 5 helices:
Transmembrane helix 1 from  60 (out) to  81 (in) :	3940.16
Transmembrane helix 2 from  99 (in) to 119 (out) :	3253.82
Transmembrane helix 3 from 190 (out) to 205 (in) :	-519.582
Transmembrane helix 4 from 209 (in) to 224 (out) :	-569.823
Transmembrane helix 5 from 253 (out) to 268 (in) :	-559.528
Score = -100000.000000

Processing 4 helices:
Transmembrane helix 1 from  19 (in) to  34 (out) :	-445.657
Transmembrane helix 2 from  60 (out) to  81 (in) :	3940.16
Transmembrane helix 3 from  99 (in) to 119 (out) :	3253.82
Transmembrane helix 4 from 193 (out) to 208 (in) :	-506.732
Score = -100000.000000

Processing 3 helices:
Transmembrane helix 1 from  60 (out) to  81 (in) :	3940.16
Transmembrane helix 2 from  99 (in) to 119 (out) :	3253.82
Transmembrane helix 3 from 193 (out) to 208 (in) :	-506.732
Score = -100000.000000

Processing 2 helices:
Transmembrane helix 1 from  60 (in) to  81 (out) :	3903.24
Transmembrane helix 2 from  99 (out) to 119 (in) :	3250.58
Score = 7.010628

Summary of topology analysis:
 1 helix   (+) : Score = 3.77255
 1 helix   (-) : Score = 4.07085
 2 helices (+) : Score = 7.01063
 2 helices (-) : Score = 7.33717
 3 helices (+) : Score = -100000
 3 helices (-) : Score = -100000
 4 helices (+) : Score = -100000
 4 helices (-) : Score = -100000
 5 helices (+) : Score = -100000
 5 helices (-) : Score = -100000

...

Results:
Signal peptide:		1-13
Signal score:		22.738
Topology:		60-81,99-119
Re-entrant helices:	Not detected.
Helix count:		2
N-terminal:		out
Score:			7.33717


All the possible topologies, and associated scores, are listed from the top
of the file. At the bottom is a summary. Topologies with weakly predicted
helices will be scored down (to -100000), as will topologies that do not
fit constraints when a constrained prediction is made. The highest scoring
topology is returned at the bottom; in this case a 2 helix prediction with
the N-terminal outside, and a predicted signal peptide. The larger the
difference in score between the highest and second highest scoring
topologies, the greater the prediction confidence.This topology is
illustrated in the files 1M57_H_schematic.png 1M57_H_cartoon_memsat_svm.png.
A signal peptide score about 8.5 will force the N-terminal to be extracellular
and will results in prediction of a signal peptide, regardless of the prior
topology prediction.

Please note that the maximum sequence length is currently set to 4000 residues,
as sequences longer than this are likely to be multi-domain proteins. The
maximum number of helices that can be predicted is currently set to 25. Both
these values can be modified by changing the values MAXSEQLEN and MAXNHEL at
the top of the memsat-svm.cpp file. However, the program was benchmarked with
the default settings so prediction accuracy may vary if you adjust these
values.



FINALLY
=======

If you  need  assistance in  getting  MEMSAT-SVM  working, or if you find any
bugs, please contact the author at the following e-mail address:

Timothy Nugent
E-mail: [email protected]

Bioinformatics Unit
Dept. of Computer Science
University College
Gower Street
London
WC1E 6BT