A bioinformatic pipeline for proteome annotation to predict if a protein is exposed on the surface of a bacteria.
inmembrane is a pipeline for proteome annotation to predict if a protein is exposed on the surface of a bacteria. It orchestrates the analysis of protein sequences to provides a summary of which targets may be surface exposed based on predicted subcellular localization signals and membrane topology. Currently protocols have been implemented for gram+ and gram- bacterial proteomes.
Typical usage is via the script inmembrane_scan, eg:
$ inmembrane_scan mysequences.fasta
The provided sequences (in FASTA format) are subjected to an number of sequence analyses using external programs (see below) and the result summarized like:
SPy_0008 CYTOPLASM(non-PSE) . SPy_0008 from AE004092 SPy_0010 PSE-Membrane tmhmm(1) SPy_0010 from AE004092 SPy_0012 PSE-Cellwall hmm(GW2|GW3|GW1);signalp SPy_0012 from AE004092 SPy_0016 MEMBRANE(non-PSE) tmhmm(12) SPy_0016 from AE004092 SPy_0019 SECRETED signalp SPy_0019 from AE004092
As well as output to stdout, this will generate a summary CSV file mysequences.csv``and a directory ``mysequences containing output files generated during the run.
Although inmembrane is primarily designed to be used as a stand alone program, it can also be used as a library like:
import inmembrane params = inmembrane.get_params() params['fasta'] = "input.fasta" annotations = inmembrane.process(params)
where annotations is a dictionary of the results, with protein sequence IDs as keys.
You can also test the functionality of the analysis plugins that are part of inmembrane by typing:
$ inmembrane_scan --test
This can be useful for determining which binary dependences are correctly installed, or exposing any broken / offline web services required for a particular analysis.
Installation and Configuration
The latest stable release of inmembrane can be installed via pip, or the bleeding edge from Github.
$ sudo pip install inmembrane
Or from Github:
$ git clone http://github.com/boscoh/inmembrane.git $ cd inmembrane $ sudo python setup.py install
The package includes tests, examples, data files, docs. HMMER3 is the only required external dependency, however for large analyses (multiple proteomes) it is suggested that local versions of other analysis programs are installed rather than relying on web services (see Installing dependencies below).
The editable parameters of inmembrane are found in inmembrane.config, which is always located in the same directory as the main script. If no such file exists, a default inmembrane.config will be generated. By default, you probably don’t need to change anything.
The parameters are:
- the path location of the binaries for SignalP, LipoP, TMHMM, HMMSEARCH, and MEMSAT. This can be the full path, or just the binary name if it is on the system path environment. Use which to check.
- ‘protocol’ to indicate which analysis you want to use.
Currently, we support:
- gram_pos the analysis of surface-exposed proteins of Gram+ bacteria;
- gram_neg annotation of subcellular localization and inner membrane topology classification for Gram- bacteria
- ‘hmm_profiles_dir’: the location of the HMMER profiles for any HMM peptide sequence motifs
- for HMMER, you can set the cutoffs for significance, the E-value ‘hmm_evalue_max’, and the score ‘hmm_score_min’
- the shortest length of a loop that sticks out of the peptidoglycan layer of a Gram+ bacteria. The SurfG+ determined this to be 50 amino acids for terminal loops, and twice that for internal loops, 100
- ‘helix_programs’ you can choose which of the transmembrane-helix prediction programs you want to use
The output of inmembrane gram_pos protocol consists of four columns of output. This is printed to stdout and written as a CSV file, which can be opened in spreadsheet software such as EXCEL. The standard text output can be parsed using space delimiters (empty fields in the third column are indicated with a “.”). Logging information are prefaced by a ‘#’ character, and is sent to stderr.
Here’s an example:
SPy_0008 CYTOPLASM(non-PSE) . SPy_0008 from AE004092 SPy_0009 CYTOPLASM(non-PSE) . SPy_0009 from AE004092 SPy_0010 PSE-Membrane tmhmm(1) SPy_0010 from AE004092 SPy_0012 PSE-Cellwall hmm(GW2|GW3|GW1);signalp SPy_0012 from AE004092 SPy_0013 PSE-Membrane tmhmm(1) SPy_0013 from AE004092 SPy_0015 PSE-Membrane tmhmm(2) SPy_0015 from AE004092 SPy_0016 MEMBRANE(non-PSE) tmhmm(12) SPy_0016 from AE004092 SPy_0019 SECRETED signalp SPy_0019 from AE004092
- the first column is the SeqID which is the first token in the identifier line of the sequence in the FASTA file
- the second column is the prediction, it is CYTOPLASM(non-PSE), MEMBRANE(non-PSE), PSE-Cellwall, PSE-Membrane, PSE-Lipoprotein or SECRETED. Any ‘PSE’ (Potentially Surface Exposed) annotation means that based on the predicted topology, the protein is likely to be surface exposed and will be protease accessible in a membrane-shaving experiment.
- the third line is a summary of features detected by external
- tmhmm(2) means 2 transmembrane helices were found by TMHMM
- hmm(GW2|GW3|GW1) means that the GW1, GW2 and GW3 motifs were found by HMMER hmmsearch
- signalp means a secretion signal was found SignalP
- lipop means a Sp II secretion signal found by LipoP with an appropriate CYS residue at the cleavage site, which will be attached to a phospholipid in the membrane
- the rest of the line gives the full identifier of the sequence in the FASTA file.
While inmembrane only requires a local installation of HMMER 3.0 and can used web services for TMHMM, SignalP, LipoP and various OMP beta-barrel predictors, for large scale analyses (5000 sequences+) it is suggested that locally installed versions are used in the interest of speed, at to be polite to publically available web services.
With each dependency, it is important that you have the exact version that inmembrane is written to interoperate with, otherwise inmembrane is likely to be unable to interpret the output of the downstream analysis program.
Required dependencies, and their versions:
- HMMER 3.0
- TMHMM 2.0 or MEMSAT3
- SignalP 4.0
- LipoP 1.0
These instructions have been tailored for Debian-based systems, in particular Ubuntu 11.10+. Each of these dependencies are licensed free to academic users.
On Ubuntu (and other Debian-derived) Linux distributions:
$ sudo apt-get install hmmer
should be enough.
- Download HMMER 3.0 from http://hmmer.janelia.org/software.
- The HMMER user guide describes how to install it. For the pre-compiled packages, this is as simple as putting the binaries on your PATH.
Only one of TMHMM or MEMSAT3 are required, but users that want to compare transmembrane segment predictions can install both.
- Download and install TMHMM 2.0 from http://www.cbs.dtu.dk/cgi-bin/nph-sw_request?tmhmm.
- In the bin/tmhmm script, edit the $opt_basedir variable to point to the full path of where TMHMM is installed.
- Download SignalP 4.0 http://www.cbs.dtu.dk/cgi-bin/nph-sw_request?signalp. You will need to fill out the form with an institutional email address and accept the academic license. The software will be emailed to you.
- Follow the installation instructions at http://www.cbs.dtu.dk/services/doc/signalp-4.0.readme.
- Download LipoP 1.0 from http://www.cbs.dtu.dk/cgi-bin/nph-sw_request?lipop. The installation proceedure is similar to that for SignalP.
- Download MEMSAT3 from http://bioinfadmin.cs.ucl.ac.uk/downloads/memsat/memsat3/ (only memsat3_academic.tar is required).
- MEMSAT3 requires NCBI BLAST (“legacy” BLAST, not BLAST+) using the SwissProt (swissprot) database.
- Legacy BLAST can be downloaded at ftp://ftp.ncbi.nlm.nih.gov/blast/executables/release/LATEST/ installed using the instructions provided by NCBI http://www.ncbi.nlm.nih.gov/staff/tao/URLAPI/unix_setup.html. We have tested with version 2.2.25.
- You will need both the ‘nr’ database and the ‘swissprot’ database, since ‘swissprot’ is indexed against ‘nr’. (The other option is to download the FASTA version of Uniprot/Swiss-Prot from ftp://ftp.uniprot.org/pub/databases/uniprot/current_release/knowledgebase/complete/uniprot_sprot.fasta.gz and create your own BLAST formatted database with using the BLAST formatdb tool).
- Edit the runmemsat script included with MEMSAT3 to point to the correct locations using absolute paths:
- ‘dbname’ is the location of your BLAST formatted swissprot database
- ‘ncbidir’ is the base directory of your BLAST installation
- ‘execdir’ is the path where the MEMSAT3 executable resides
- ‘datadir’ is the the path to the MEMSAT3 data directory )
(Note the the ‘runmemsat’ script refers to PSIPRED v2, but it means MEMSAT3 - PSIPRED is NOT required).
It is a fact of life for bioinformatics that new versions of basic tools changes output formats and API. We believe that it is an essential skill to rewrite parsers to handle the subtle but significant changes in different versions. We have written inmembrane to be easily modifiable and extensible. Protocols which embody a particular high level workflow are found in inmembrane/protocols.
All interaction with a specific external programs or web services have been wrapped into a single python plugin module, and placed in the inmembrane/plugins directory. This contains the code to both run the program and to parse the output. We have tried to make the parsing code as concise as possible. Specifically, by using the native Python dictionary, which allows an enormous amout of flexibility, we can collate the results of various analyses with very little code.
A more comprehensive overview can be found at http://boscoh.github.com/inmembrane/api.html.
inmembrane development style guide:
Here are some guidelines for understanding and extending the code.
- Confidence: Plugins that wrap an external program should have at least one high level test which is executed by run_tests.py. This allows new users to immediately determine if their dependencies are operating as expected.
- Interface: A plugin that wraps an external program must receive a params data structure (derived from inmembrane.config) and a proteins data structure (which is a dictionary keyed by sequence id). Plugins should return a ‘proteins’ object.
- Flexibility: Plugins should have a ‘force’ boolean argument that will force the analysis to re-run and overwrite output files.
- Efficiency: All plugins should write an output file which is read upon invocation to avoid the analysis being re-run.
- Documentation: A plugin must have a Python docstring describing what it does, what parameters it requires in the params dictionary and what it adds to the proteins data structure. See the code for examples.
- Anal: Unique sequence ID strings (eg gi|1234567) are called ‘seqid’. ‘name’ is ambiguous. ‘prot_id’ is reasonable, however conceptually a ‘protein’ is not the same thing as a string that represents it’s ‘sequence’ - hence the preference for ‘seqid’.
- Anal: All file handles should be closed when they are no longer needed.