PSIPRED RELEASE NOTES
=====================

PSIPRED Version 4.0

By David Jones, January 2016

*** IMPORTANT *****************************************************
NCBI are now trying to move users to the new BLAST+ package. Please
see the README file in the BLAST+ subdirectory for more information
on PSIPRED's support for BLAST+. For now the preferred option is
to stick with the classic BLAST package as the default. If the tar
or rpm file you are downloading from NCBI has "+" in the filename,
then you are downloading BLAST+ rather than BLAST.
*******************************************************************

Here are some very brief notes on using the PSIPRED V4 software.

PSIPRED is supplied in source code form - it must be compiled before
it can be used. The code should compile on any ANSI C compiler e.g.
the GNU C compiler.

Please see the LICENSE file for the license terms for the software.
Basically it's free to anyone (including commercial users) as long as
you don't want to sell the software or, for example, store the results
obtained with it in a database and then try to sell the database.
If you do wish to sell the software or use it in a commercial product,
then please contact UCL Business (http://www.uclb.com).

PSIPRED is run via a tcsh shell script called "runpsipred" - this is a
very simple script which you should be able to convert to Perl or whatever
scripting language you like.

If your sequence does not have any homologues in the current data banks,
then it is possible to run PSIPRED on a single sequence. In this case,
PSIPRED is run via a tcsh shell script called "runpsipred_single". Unfortunately,
like every other secondary structure prediction method, PSIPRED does not
perform as well on single sequences. Any secondary structure prediction based on
a single sequence should be considered as unreliable.

Before running PSIPRED, please check the runpsipred and runpsipred_single scripts
to see if the path variables are set to wherever you have installed the
program and data files. The default is to assume that the program is
installed in the current directory - this is probably NOT what you want!

INSTALLATION VIA GITHUB AND ANSIBLE
===================================

First ensure that ansible is installed on your system, then clone the github
repo

% pip install ansible
% git clone https://github.com/psipred/psipred.git
% cd ~/psipred/ansible_installer

Next edit the the config_vars.yml to reflect where you would like psipred and 
its underlying data to be installed. You can choose a version of uniref but
we recommend uniref90 for most accurate results.

You can now run ansible as per

% ansible-playbook -i hosts install.yml

You can edit the hosts file to install psipred on one or more machines.

MANUAL INSTALLATION
===================

Firstly compile the software:

tcsh% cd to-wherever-you-untarred-PSIPRED

tcsh% cd src

tcsh% make

tcsh% make install

The executables will be placed in the PSIPRED bin directory.

You must also install the PSI-BLAST and Impala software from the
NCBI toolkit, and also install appropriate sequence data banks.

The NCBI toolkit can be obtained from URL ftp://ftp.ncbi.nih.gov

PSI-BLAST executables can be obtained from ftp://ftp.ncbi.nih.gov/blast



EXAMPLE USAGE
=============

In this example the target sequence is called "example.fasta":

tcsh% runpsipred example.fasta

Running PSI-BLAST with sequence example.fasta ...
Predicting secondary structure...
Pass1 ...
Pass2 ...
Cleaning up ...
Final output file: example.horiz
Finished.

That's it - you can then look at the output:

tcsh% more example.horiz


SPECIAL OPTIONS
===============

The psipass2 program has several special options which you can use if you wish.

For example, the default command is as follows:

psipass2 weights_p2.dat 1 1.0 1.0 output.ss2 input.ss > output.horiz

Arguments 2,3 & 4 are as follows:

Argument 2: No of filter iterations
This controls the amount of "smoothing" that is carried out on the final
prediction. The recommended setting is 1, but it may be worth trying
higher values to increase the level of smoothing.

Argument 3&4: Helix/Strand Decision constants
These options control the bias for helix (Arg3) and strand (Arg4) predictions.
The default values are equal to 1.0, but if you know your protein is, for
example, mostly comprised of beta strands then you can increase the bias
towards beta strand prediction. For example:

psipass2 weights_p2.dat 1 1.0 1.3 output.ss2 input.ss > output.horiz

increases the bias towards beta strand prediction by approximately 30%.

 

SEQUENCE DATA BANK
==================

As of PSIPRED V4.0 onwards, we no longer believe it is necessary for the sequence
data banks used with PSI-BLAST to be filtered to remove low-complexity regions,
transmembrane regions, and coiled-coil segments. The search data bank can
therefore be any large non-redundant protein sequence data bank, with
UNIREF90 (http://www.uniprot.org/help/uniref) being the recommended one.



CHANGES FROM THE ORIGINAL PSIPRED
=================================

The following is a quick summary of the main changes since the original
PSIPRED.

1. The program now makes use of PSI-BLAST binary checkpoint files (using the
Impala program makemat) to reduce loss of precision when parsing the original
ASCII position specific matrices.

2. By default the 1st pass uses an average of 3 different neural network
weight sets - this improves prediction accuracy slightly.

3. In addition to the normal horizontal summary output format, the program
now also produces a full table of results which shows the individual
coil, helix, strand network outputs.

4. A one-line header is output at the start of the output files to allow
THREADER (and other programs) to automatically recognise a PSIPRED
prediction.

5. An experimental interface to BLAST+ has been added (V3.0). This will extract
PSSM data directly from ASN.1 checkpoint files.

6. Minor formatting bugs in .horiz file output for very long sequences
have now been fixed (V3.21).

7. Minor output bug loses singleton residue coil predictions fixed (V3.3)

8. V4.0 released: new neural network architectures.