- Article -
| In Silico Biology 8, 0013 (2008); ©2008, Bioinformation Systems e.V. |
Vector-G: Multi-modular SVM-based heterotrimeric G protein prediction
Preti Jain1, Puneet Wadhwa2, Ramazan Aygun2* and Gopi Podila1
1 Department of Biological Sciences, University of Alabama in Huntsville, Huntsville AL, 35899, USA
2 Computer Science Department, University of Alabama in Huntsville, Huntsville AL, 35899, USA
URL: http://www.davvbiotech.res.in
* Corresponding author
Email: raygun@cs.uah.edu
Edited by H. Michael; received October 13, 2007; revised February 14, 2008; accepted February 15, 2008; published April 12, 2008
Abstract
Heterotrimeric G proteins interact with G protein-coupled receptors in response to stimulation by hormones, neurotransmitters, chemokines, and sensory signals to intracellular signaling cascades. Recently reported studies indicate that G protein subunits play a significant role in different eukaryotic diseases including inflammation, neurological diseases, cardiovascular diseases, endocrine disorders as well as plant pathogen response, infectious hyphae growth, differentiation and virulence of pathogenic fungi. Thus a study of their functions, signaling pathways, and protein interactions may lead to the development of various preventive approaches. The diversity of α, β and γ subunits of G proteins necessitates a prediction algorithm that helps in the identification of new proteins such as Gβ where WD-40 repeats are not well characterized. The currently available techniques for finding G proteins are homology based search analyses and wet lab experiments, which are not very effective in finding new classes of proteins. We present here a robust computational method for finding new G proteins and their homologs using a SVM based pattern recognition algorithm. Several physicochemical and compositional properties including dipeptide, tripeptide and hydrophobicity composition are used for generating the SVM classifiers. This method has 96.17%, 95.38%, 97.6% sensitivity and 99.45%, 100%, 100% specificity on test sets for G protein α, β, and γ subunits, respectively. This algorithm correctly predicts the known α, β and γ subunits reported in literature. One important contribution of this algorithm is that it helps in improving genome annotation of several proteins as G proteins and serves as a useful tool for comparative genomic analysis of G proteins. Using this method, novel G protein subunits are predicted in 31 genomes covering plant, fungi and animal kingdom. The software is available at the website http://biomine.cs.uah.edu/bioinformatics/svm_prog/scripts/GProteins/vectorg.html.
Keywords: heterotrimeric G proteins, SVM, compositional properties, signal transduction