BETAWARE version 1.0
Copyright (C) 2012 Castrense Savojardo
BETAWARE is a software package designed for the analysis of trans-membrane beta-barrel proteins. Basically, it offers two different functionalities:
- Detection of beta-barrel proteins
- Prediction of trans-membrane topology
BETAWARE is based on machine-learning methods. The detection is performed using a predictor based on N-to-1 Extreme Learning Machines (ELMs). See ref [1] for more details about the method. Topology prediction is carried out using Grammatical-Restrained Conditional Random Fields (GRHCRFs) (ref [2]).
Proteins should be provided to BETAWARE in the form of sequence profiles. Given a protein sequence of length L, a profile is a position specific Lx20 matrix whose component (i,a) represent the relative frequency of amino acid type a at position i computed from a multiple sequence alignment (MSA). The MSA is usually obtained from the protein homologous sequences found using BLAST or PSI-BLAST against a non-redundant database of sequences. We recommend using PSI-BLAST since this software has been adopted to generate sequence profiles used to train BETAWARE.
BETAWARE is entirely written in Python and designed to run on Unix/Linux systems. BETAWARE assumes the following software packages are installed in your system:
- python v2.6
- python argparse library (http://docs.python.org/library/argparse.html#module-argparse) This library is used to parse command line options.
- python numpy and scipy libraries (http://numpy.scipy.org/)
To install these packages under Linux debian/ubuntu (you need to be a superuser):
sudo apt-get install python-numpy python-scipy python-argparse
As mentioned above, BETAWARE can be used to detect a beta-barrel protein and to predict its topology. You need to provide BETAWARE with the protein sequence in FASTA format and the sequence profile.
Once the program tarball has been downloaded and uncompressed the root directory looks like the following:
betaware/
bin/
betaware.py
data/
...
example/
...
modules/
...
predBeta.sh
test.sh
LICENCE
README
To use BETAWARE, from the package root, run:
./predBeta.sh FASTA_FILE PROFILE_FILE
This script runs BETAWARE with default options and prints the result in the standard output. The example directory contains some examples files useful to become familiar with the program. In the following example, BETAWARE runs on the example protein 1qj8_A:
./predBeta.sh example/1qj8.fasta example/1qj8.prof
The output would be:
Sequence id : 1QJ8:A|PDBID|CHAIN|SEQUENCE
Sequence length : 148
Predicted TMBB : Yes
Topology : 2-9,23-29,35-46,60-69,78-87,103-114,120-131,135-145
Seq : ATSTVTGGYAQSDAQGQMNKMGGFNLKYRYEEDNSPLGVIGSFTYTEKSRTASSGDYNKN
SS : iTTTTTTTToooooooooooooTTTTTTTiiiiiTTTTTTTTTTTToooooooooooooT
Prob: cbaaaaaaaaaaaaaaaaaaccaaaaaaacaaabccaaaaaaabbdaaaaaaaaaacdeb
------------------------------------------------------------------
Seq : QYYGITAGPAYRINDWASIYGVVGVGYGKFQTTEYPTYKNDTSDYGFSYGAGLQFNPMEN
SS : TTTTTTTTTiiiiiiiiTTTTTTTTTToooooooooooooooTTTTTTTTTTTTiiiiiT
Prob: aaaaaaaaabaaaaaabaaaaaaaaaaaaaaaaaaaaaaaaadaaaaaaaaabdaaaaae
------------------------------------------------------------------
Seq : VALDFSYEQSRIRSVDVGTWIAGVGYRF
SS : TTTTTTTTTTToooTTTTTTTTTTTiii
Prob: baaaaaaaaaaaaaaaaaaaaaaaabaa
----------------------------------
//
where the row labeled with "Prob:" reports the posterior probability of the label at each position (from a="probability close to 1" to j="probability close to 0").
BETAWARE can be run with more advanced options other than the default ones. To do this you should refer the main Python script betaware.py which can be found into the bin/ directory.
In order to run betaware.py you first need to set the environment variable BETAWARE_ROOT to point to the root directory of the package. This can be done using the following command:
export BETAWARE_ROOT=/path/to/betware/root/directory
For example, if betaware has been uncompressed into /home/cas/betaware you should run:
export BETAWARE_ROOT=/home/cas/betaware
** IMPORTANT NOTICE ** In principle the BETAWARE_ROOT variable should be exported EVERY TIME you open a new terminal. To avoid this you simply need to cut and paste the command above into a bash shell startup file such as ~/.profile. **
Once BETAWARE_ROOT has been exported you would be able to run the program. You may also want to add $BETAWARE_ROOT/bin into you PATH or put $BETAWARE_ROOT/bin/betaware.py in some directory already listed in your PATH.
The script test.sh run BETAWARE on the example data with different options.
In the first example, BETAWARE is used to detect a beta-barrel in the
protein 1qj8_A and predict its topology:
betaware.py -f example/1qj8.fasta -p example/1qj8.prof
The -f option is used to specify the path of the protein sequence in FASTA format. The sequence profile file is provided to the program with the -p option. The output of the command above will look like the following:
Sequence id : 1QJ8:A|PDBID|CHAIN|SEQUENCE
Sequence length : 148
Predicted TMBB : Yes
Topology : 2-9,23-29,35-46,60-69,78-87,103-114,120-131,135-145
Seq : ATSTVTGGYAQSDAQGQMNKMGGFNLKYRYEEDNSPLGVIGSFTYTEKSRTASSGDYNKN
SS : iTTTTTTTToooooooooooooTTTTTTTiiiiiTTTTTTTTTTTToooooooooooooT
Prob: cbaaaaaaaaaaaaaaaaaaccaaaaaaacaaabccaaaaaaabbdaaaaaaaaaacdeb
------------------------------------------------------------------
Seq : QYYGITAGPAYRINDWASIYGVVGVGYGKFQTTEYPTYKNDTSDYGFSYGAGLQFNPMEN
SS : TTTTTTTTTiiiiiiiiTTTTTTTTTToooooooooooooooTTTTTTTTTTTTiiiiiT
Prob: aaaaaaaaabaaaaaabaaaaaaaaaaaaaaaaaaaaaaaaadaaaaaaaaabdaaaaae
------------------------------------------------------------------
Seq : VALDFSYEQSRIRSVDVGTWIAGVGYRF
SS : TTTTTTTTTTToooTTTTTTTTTTTiii
Prob: baaaaaaaaaaaaaaaaaaaaaaaabaa
----------------------------------
//
The topology prediction is performed and reported only if the protein is predicted as trans-membrane beta-barrel (TMBB). However, you may want to predict the topology also when the protein is not identified as TMBB. In the following example the -t option tells the program to always report the topology:
betaware.py -t -f example/12ca.fasta -p example/12ca.prof
The output would be:
Sequence id : 12CA:A|PDBID|CHAIN|SEQUENCE
Sequence length : 260
Predicted TMBB : No
TMB Strands : 58-64,71-79,99-106,116-124
Seq : MSHHWGYGKHNGPEHWHKDFPIAKGERQSPVDIDTHTAKYDPSLKPLSVSYDQATSLRIL
SS : iiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiTTT
Prob: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaabbcccccccbbc
------------------------------------------------------------------
Seq : NNGHAFNVEFDDSQDKAVLKGGPLDGTYRLIQFHFHWGSLDGQGSEHTVDKKKYAAELHL
SS : TTTTooooooTTTTTTTTTiiiiiiiiiiiiiiiiiiiTTTTTTTToooooooooTTTTT
Prob: ccccaaaaaaedccccbbdcccccccabbaaaaaaaaaaaaaaaaaddbaabbcdbbaaa
------------------------------------------------------------------
Seq : AHWNTKYGDFGKAVQQPDGLAVLGIFLKVGSAKPGLQKVVDVLDSIKTKGKSADFTNFDP
SS : TTTTiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii
Prob: accdbaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
------------------------------------------------------------------
Seq : RGLLPESLDYWTYPGSLTTPPLLECVTWIVLKEPISVSSEQVLKFRKLNFNGEGEPEELM
SS : iiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii
Prob: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
------------------------------------------------------------------
Seq : VDNWRPAQPLKNRQIKASFK
SS : iiiiiiiiiiiiiiiiiiii
Prob: aaaaaaaaaaaaaaaaaaaa
--------------------------
//
As you can see the protein 12ca_A is not predicted as TMBB. However its topology has been predicted and reported.
BETAWARE prints the output to stdout by default. You can change this behavior using -o option:
betaware.py -o example/1qj8.out -f example/1qj8.fasta -p example/1qj8.prof
The -a option is used to specify the correspondence between amino acids and columns in the sequence profile file. BETAWARE machine-learning algorithms have been trained using the following correspondence:
VLIMFWYGAPSTCHRKQEND
In your profile, columns may be arranged in a different way. With the -a option you tell BETAWARE which column corresponds to which amino acid and let the program rearrange the matrix properly. For instance, if your columns are sorted in alphabetic order you would run:
betaware.py -a ACDEFGHIKLMNPQRSTVWY -f example/1qj8.fasta -p example/1qj8ALPH.prof
and the output would be:
Sequence id : 1QJ8:A|PDBID|CHAIN|SEQUENCE
Sequence length : 148
Predicted TMBB : Yes
Topology : 2-9,23-29,35-46,60-69,78-87,103-114,120-131,135-145
Seq : ATSTVTGGYAQSDAQGQMNKMGGFNLKYRYEEDNSPLGVIGSFTYTEKSRTASSGDYNKN
SS : iTTTTTTTToooooooooooooTTTTTTTiiiiiTTTTTTTTTTTToooooooooooooT
Prob: cbaaaaaaaaaaaaaaaaaaccaaaaaaacaaabccaaaaaaabbdaaaaaaaaaacdeb
------------------------------------------------------------------
Seq : QYYGITAGPAYRINDWASIYGVVGVGYGKFQTTEYPTYKNDTSDYGFSYGAGLQFNPMEN
SS : TTTTTTTTTiiiiiiiiTTTTTTTTTToooooooooooooooTTTTTTTTTTTTiiiiiT
Prob: aaaaaaaaabaaaaaabaaaaaaaaaaaaaaaaaaaaaaaaadaaaaaaaaabdaaaaae
------------------------------------------------------------------
Seq : VALDFSYEQSRIRSVDVGTWIAGVGYRF
SS : TTTTTTTTTTToooTTTTTTTTTTTiii
Prob: baaaaaaaaaaaaaaaaaaaaaaaabaa
----------------------------------
//
which is identical to the output of the first example.
Finally, with the option -s it is possible to adjust the sensitivity of the detection algorithm. The sensitivity can be specified through a float value between 0 and 1. The higher is the sensitivity the smaller would be the chance to obtain false negatives. However, a high sensitivity also increases the probability of getting false positives. By default, this parameter is set to 0.5. For instance, by reducing the sensitivity the 1qj8_A protein will be predicted as non-TMBB:
betaware.py -s 0.05 -f example/1qj8.fasta -p example/1qj8.prof
while the 12ca_A can be predicted as TMBB by increasing the sensitivity:
betaware.py -s 0.95 -f example/12ca.fasta -p example/12ca.prof
If not set properly this parameter can seriously affect the performance of the detection algorithm. We recommend to maintain the sensitivity around 0.5.
[1] Savojardo C., Fariselli P., Casadio R., Improving the detection of transmembrane beta-barrel chains with N-to-1 Extreme Learning Machines, Bioinformatics 27 (22): 3123-3128, 2011. [2] Fariselli P., Savojardo C., Martelli P.L., Casadio R., Grammatical-Restrained Hidden Conditional Random Fields for Bioinformatics Applications, AlMoB 4:13, 2009.