Efficient Spritz (ESpritz)
--------------------------------------------------------------
 (c)  Ian Walsh and Silvio C.E. Tosatto,  2011.
       E-mail:   <ian.walsh@bio.unipd.it>
       --------------------------------------------------------------
       README File
       Version 1.1

       Espritz - Efficient prediciton of protein disorder segments
--------------------------------------------------------------

1. Introduction
Espritz is software to detect regions of proteins which are thought to contain no structural content. This non-structure
is known as protein disorder. Espritz uses an efficient and accurate prediction algorithm based on Bi-directional 
Recursive Neural networks (BRNN's). BRNN's are a sequence based machine learning algorithm which have being found to be useful 
for other structural predictions such as secondary structure. The method is based solely on the sequence and does not use 
expensive computations to find multiple sequence alignments.
For a slightly more detailed technical description of the capabilities and technology used, see the methods page on our server 
(http://protein.bio.unipd.it/espritz/). For the most detailed description see our "I. Walsh, A.J.M Martin and S. Tosatto. 
Sequence is sufficient for accurate disorder predictions applied to various genomes, Bioinformatics, SUBMITTED, 2011". 

INSTALL

	***** This software has been tested only on linux based systems. Perl, C++ runtime must be installed on your system 
	***** This software may be installed on MAC OS and Windows but we have no support at the moment for these operating systems. 



	The sequence based fast predictors should work without any tweaking. This is because all files are contained within your folder. 
	

		
	However for psi-blast based predictions we recommend the following procedure: 

	(1) Download the latest BLAST binaries at ftp://ftp.ncbi.nlm.nih.gov/blast/executables/release/LATEST/. We
	used blast-2.2.25-ia32-linux.tar.gz (36MB) for our linux distribution. 
	(2) Deflate the compressed file: 
		tar -xzf blast-2.2.25-ia32-linux.tar.gz

	(3) In your new installation record the path to the blastlocation/blast-2.2.25/bin/ directory where blastlocation 
	is the directory where you decompressed the blast-2.2.25-ia32-linux.tar.gz file. 
	
	(4) Edit the ./align/getAlignments.pl perl script and replace the $blastdir variable with your new 
	path blastlocation/blast-2.2.25/bin/
	
	(4) Next we need to download the latest blast databases. Conveniently the newest blast distribution above 
	contains a script to update your sequence databases. Execute the following:
		cd blastlocation/ncbi-blast-2.2.25+/
		mkdir blastdb
		download nr.gz at ftp://ftp.ncbi.nih.gov/blast/db/FASTA/
		
	Decompress the latest nr sequence database: 
		gzip -d nr.gz

	(5) The sequences contain a lot of redundancy. Filter the sequences with the "cd-hit" method http://www.bioinformatics.org/cd-hit/. 
	Once it is downloaded decompress it:
		cd location_of_cd-hit
		tar -xzf cd-hit*
		make
		sudo make install (or do "make install" as administrator)
		cdhit -c 0.9 -i nr -o nr90

	(6) Our tests have shown that filtering the fasta sequences downloaded in (4) with masked low complexity regions were useful 
	for our disorder predictors. For this we used the pfilt agorithm http://bioinfadmin.cs.ucl.ac.uk/downloads/pfilt/. Follow the instructions 
	at this site to compile pfilt. Then execute: 
		./pfilt nr90 > nr90filt
	The resulting nr90filt are the final sequences used for calculations of alignments. Alignments generated in this manner were found to 
	be optimal for disorder prediction. 

	(7) Format the nr90filt into the blastdb format: 
		blastlocation/blast-2.2.25/bin/formatdb -p T -i nr90filt

	(8) Finally, create a file called .ncbirc on your home directory (e.g. /home/user/) and add:
		[NCBI]
		data=/home/user/blast/data

		[BLAST]
		BLASTDB=/home/user/blast/db
		BLASTMAT=/home/user/blast/data
	Save and start using ESpritz in a new terminal. 

Notes:
* (1) and (4) may be time consuming depending on your internet connection. Moreover, several gigabytes of hard disk space will be used.  

* The above psi-blast setup is only a recommendation. Your favorite method can be used. However, be careful that 
the script ./align/getAlignments.pl uses the "psiblast" binary which replaces the old "blastpgp" binary. 

* We recommend not to use psi-blast unless you want the extra 1-2% accuracy. We believe the advatages of just using the sequence based predictors outway
the advantages for using ESpritz with psi-blast. 



Input Options
			Working directory
	This is the directory which contains all the fasta files which ESpritz will predict. IMPORTANT: The files must have the .fasta extension.
	See EXAMPLES.sh to see how to run the software on some example fasta sequences.

			x-ray, Disprot or NMR disorder
	* x-ray: This is based on missing atoms from the Protein Data Bank (PDB) X-ray solved structures. If this option is 
	chosen then the predictors with short x-ray disorder options are executed. See EXAMPLES.sh to see how to select this option.
	* disprot: The dataset used for this definition contains long disorder segments. In particular, disprot a manually curated 
	database which is often based on functional attributes of the disordered region. If this option is chosen then the predictors 
	with long disorder options are executed. See EXAMPLES.sh to see how to select this option.
	* nmr: The dataset used for this definition contains was defined using our server MOBI (http://protein.bio.unipd.it/mobi). 
	This disorder is basedon the flexibility/mobility between different models in NMR structures. See EXAMPLES.sh to see how to select this option.


			Probability decision
	*  Best Sw: The definition of Sw can be found in the CASP assessments. This option chooses the threshold which maximised the 
	Sw measure on the training set. This threshold tends to over-predict disorder but is nonetheless useful. See EXAMPLES.sh to 
	see how to select this option.
        * 5% False Positive Rate (FPR): This is a stricter definition of disorder compared to best Sw. It is the threshold which 
	produces 5% False positives on the training data. See EXAMPLES.sh to see how to select this option.


Output
	After prediction all the fasta files in the working directory will have its corresponding prediction file (.espritz extension). These prediction 
	files will also be placed in the working directory choosen at input.



References: 

Please cite this if you use our software):
Espritz bla bla

psi-blast
Altschul, S.F., et al. (1997) Gapped BLAST and PSI-BLAST: a new generation of
protein database search programs, Nucleic Acids Res, 25, 3389-3402.


cd-hit
"Clustering of highly homologous sequences to reduce the size of large protein database", Weizhong Li, Lukasz 
Jaroszewski & Adam Godzik Bioinformatics, (2001) 17:282-283

