Introduction

This repository houses a Python 3.5+ implementation of transmembrane helix hidden Markov model (TMHMM) originally described in:

E. L.L. Sonnhammer, G. von Heijne, and A. Krogh. A hidden Markov model for predicting transmembrane helices in protein sequences. In J. Glasgow, T. Littlejohn, F. Major, R. Lathrop, D. Sankoff, and C. Sensen, editors, Proceedings of the Sixth International Conference on Intelligent Systems for Molecular Biology, pages 175-182, Menlo Park, CA, 1998. AAAI Press.

Why?

I did this for a few reasons:

• the source code is not available as part of the publication,
• the downloadable binaries are Linux-only,
• the downloadable binaries may not be redistributed, so it's not possible to put them in a Docker image or a VM for other people to use,
• the need to predict transmembrane helices on a large dataset, which rules out the web service.

This Python implementation includes a parser for the undocumented file format used to describe the model and a pretty fast Cython implementation of the Viterbi algorithm used to perform the annotation. The tool will output files similar to the files produced by the original TMHMM implementation.

Incompatibilities

• The original TMHMM implementation handles ambigious characters and gaps in an undocumented way. However, tmhmm.py does not attempt to handle such characters at all and will fail. A possible fix is to replace those characters with something also based on expert/domain knowledge. For details, see issue #9.

Installation

This package supports Python 3.5, 3.6, and 3.7. Install with:

$ pip install tmhmm.py

Only Linux is supported at the moment.

Usage

$ tmhmm -h
  usage: tmhmm [-h] -f SEQUENCE_FILE [-m MODEL_FILE] [-p]

  optional arguments:
    -h, --help            show this help message and exit
    -f SEQUENCE_FILE, --file SEQUENCE_FILE
                          path to file in fasta format with sequences
    -m MODEL_FILE, --model MODEL_FILE
                          path to the model to use
    -p, --plot            plot posterior probabilies

The -p/--plot option will only be available if matplotlib is installed and importable.

Say we have the following sequence in FASTA format in a file called test.fa:

>B9DFX7|1B|HMA8_ARATH Copper-transporting ATPase PAA2, chloroplastic  [Arabidopsis thaliana ]
MASNLLRFPLPPPSSLHIRPSKFLVNRCFPRLRRSRIRRHCSRPFFLVSNSVEISTQSFESTESSIESVKSITSDTPIL
LDVSGMMCGGCVARVKSVLMSDDRVASAVVNMLTETAAVKFKPEVEVTADTAESLAKRLTESGFEAKRRVSGMGVAENV
KKWKEMVSKKEDLLVKSRNRVAFAWTLVALCCGSHTSHILHSLGIHIAHGGIWDLLHNSYVKGGLAVGALLGPGRELLF
DGIKAFGKRSPNMNSLVGLGSMAAFSISLISLVNPELEWDASFFDEPVMLLGFVLLGRSLEERAKLQASTDMNELLSLI
STQSRLVITSSDNNTPVDSVLSSDSICINVSVDDIRVGDSLLVLPGETFPVDGSVLAGRSVVDESMLTGESLPVFKEEG
CSVSAGTINWDGPLRIKASSTGSNSTISKIVRMVEDAQGNAAPVQRLADAIAGPFVYTIMSLSAMTFAFWYYVGSHIFP
DVLLNDIAGPDGDALALSLKLAVDVLVVSCPCALGLATPTAILIGTSLGAKRGYLIRGGDVLERLASIDCVALDKTGTL
TEGRPVVSGVASLGYEEQEVLKMAAAVEKTATHPIAKAIVNEAESLNLKTPETRGQLTEPGFGTLAEIDGRFVAVGSLE
WVSDRFLKKNDSSDMVKLESLLDHKLSNTSSTSRYSKTVVYVGREGEGIIGAIAISDCLRQDAEFTVARLQEKGIKTVL
LSGDREGAVATVAKNVGIKSESTNYSLSPEKKFEFISNLQSSGHRVAMVGDGINDAPSLAQADVGIALKIEAQENAASN
AASVILVRNKLSHVVDALSLAQATMSKVYQNLAWAIAYNVISIPIAAGVLLPQYDFAMTPSLSGGLMALSSIFVVSNSL
LLQLHKSETSKNSL

We can then run tmhmm.py on this file using the following command:

$ tmhmm -m TMHMM2.0.model -f test.fa

This produces a bunch of files. One is the summary:

$ cat B9DFX7|1B|HMA8_ARATH.summary
0-444: outside
445-467: transmembrane helix
468-820: inside
821-843: transmembrane helix
844-852: outside
853-870: transmembrane helix
871-882: inside

An annotation in FASTA format:

$ cat B9DFX7|1B|HMA8_ARATH.annotation
>B9DFX7|1B|HMA8_ARATH Copper-transporting ATPase PAA2, chloroplastic  [Arabidopsis thaliana ]
OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO
OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO
OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO
OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO
OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO
OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOMMMMMMMMMMMMMMMMMMMMMMMiiiiii
iiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii
iiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii
iiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii
iiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiii
iiiiiiiiiiiiiiiiiiiiiiiiiiiiiiiMMMMMMMMMMMMMMMMMMMMMMMoooooooooMMMMMMMMMMMMMMMM
MMiiiiiiiiiiii

And finally a file containing the posterior probabilities for each label for plotting.

$ cat B9DFX7|1B|HMA8_ARATH.plot
inside membrane outside
0.20341044516 0.0 0.79658955484
0.210104176071 2.77194446172e-08 0.78989579621
0.189291062167 3.11365191554e-08 0.810708906697
0.253334801857 7.17866017044e-07 0.746664480277
0.126185012808 1.34197873962e-05 0.873801567405
...

If the -p flag is set a plot in PDF format will also be produced, following the same naming scheme as the other output files.

API

You can also use tmhmm.py as a library:

import tmhmm
annotation, posterior = tmhmm.predict(sequence, 'mymodel.model')

This returns the annotation as a string and the posterior probabilities for each label as a numpy array with shape (len(sequence), 3) where column 0, 1 and 2 corresponds to being inside, transmembrane and outside, respectively.

If you don't need the posterior probabilities set compute_posterior=False, this will save quite a lot of computation:

annotation, posterior = tmhmm.predict(
    sequence, 'mymodel.model', compute_posterior=False
)