A method for the prediction of GPCRs coupling specificity to G-proteins using refined profile Hidden Markov Models

Using a dataset of 282 GPCR sequences of known coupling preference to G-proteins and adopting a five-fold cross-validation procedure, the method yielded an 89.7% correct classification rate. In a validation set comprised of all receptor sequences that are species homologues to GPCRs with known coupling preferences, excluding the sequences used to train the models, our method yields a correct classification rate of 91.0%. Furthermore, promiscuous coupling properties were correctly predicted for 6 of the 24 GPCRs that are known to interact with more than one subfamily of G-proteins.Our method demonstrates high correct classification rate. Unlike previously published methods performing the same task, it does not require any transmembrane topology prediction in a preceding step. A web-server for the prediction of GPCRs coupling specificity to G-proteins available for non-commercial users is located at http://bioinformatics.biol.uoa.gr/PRED-COUPLE webcite.G-protein coupled receptors are important receivers of information input to eukaryotic cells. They share a common fold of seven transmembrane helices arranged as a seven α-helix bundle, as confirmed by analysis of the crystal structure of Rhodopsin [1] that has been extensively used as template for homology-based modeling of GPCRs [2-4]. A collection of messages of extreme diversity including photons and native agonists, such as ions, odorants and pheromones, amino acids, nucleotides, peptides, biogenic amines, prostaglandines and glycoprotein hormones [5] interact with different extracellular and/or transmembrane domains of GPCRs, in order to convey their messages to the interior of the cell [2,6]. Based primarily on shared sequence motifs, six distinct families of GPCRs are traditionally defined: A, B, C, D, E and the frizzled/smoothened family, as summarized in the GPCRDB classification scheme [7]. Various methods have been deployed for higher-level classification of GPCRs including profile Hidden Markov Models [8,9]