The model


A GPCR detection method


This method predicts G protein-coupled receptors from amino acid sequence.

GPCRHMM is described in

Markus Wistrand*, Lukas Käll* and Erik L.L. Sonnhammer.
A general model of G protein-coupled receptor sequences and its application to detect remote homologs.
Protein Science, 15 (3):509-21, Mars 2006.
*These authors contributed equally to this work.


To install the method you need to:
  1. Download gpcrhmm package frome here
  2. untar it in a suitable directory using the command tar xvzf gpcrhmm.tar.gz.
  3. Obtain a copy of phobius.
  4. Copy the executable decodeanhmm from the phobius directory to the gpcrhmm directory.
  5. Make a link from to your standard path e.g /usr/local/bin.


The program takes proteins in FASTA format. It recognizes the 20 amino acids and B, Z, and X, which are all treated equally as unknown. Any other character is changed to X, so please make sure the sequences are sensible proteins.

This is an example (one protein):

>EBI2_HUMAN P32249 [Homo sapiens (human)] EBV-induced G protein-coupled receptor 2 (EBI2).

GPCR detection

The usage of the method is to detect putative GPCRs in from amino acid sequence.
Either give the name of the local file in which you have the proteins in the, or paste the sequence(s) into the input form. If you paste; make sure that the first line of each entry (the identifier) is not broken.
There is a global and a local score (explained below). The local score is only calculated for the proteins that have a global score higher than a threshold of 0. To only use global scoring is faster, but tend to accumulate false positives.

Global/Local score

GPCRHMM is based on a hidden Markov model that mimics the common topology of GPCRs: they all span the membrane sevenfold. Global scoring takes the sequence through the entire model, including the N- and C-terminal sections. The reported score is calculated using the forward algorithm.

Local scoring takes the subsequence that spans the 1st-7th TM-helix (as predicted by the 1-best algorithm using the global model). The subsequence is scored to a core model that only have compartments corresponding to the 1st-7th TM-helix (forward algorithm). The local score can detect false positives that don't fit the model well, but which have long N- or C-terminal regions that accumulate score because of biased sequence composition.


Output is global score, local score and a prediction (GPCR/no). All reported scores are log-odds scores related to a null model. The null model reflects the amino acid composition in SwissProt.