RF-SBC-based approach for discriminating G protein-coupled receptors using amino acid composition property

G protein-coupled receptors (GPCRs) constitute the largest superfamily of integral membrane proteins (IMPs) that extremely contribute in the flow of information into cells. In this paper, a random forest (RF) approach for determining the importance of input variables and the subtractive clustering and fuzzy c-means (SBC) method have been proposed for discriminating the GPCRs from non-GPCRs using twenty amino acid compositions of GPCR sequences. The studied dataset was derived from UniProt/SWISSPROT database and consists of 1000 GPCR and 1000 non-GPCR reviewed sequences. The RF-SBC-based method discriminates GPCRs and non-GPCRs successfully with the accuracy, sensitivity, specificity and Matthew‘s coefficient correlation (MCC) rates of 98.7%, 98.81%, 98.59% and 0.974, respectively. These rates were obtained from averaged values of 5-fold cross validation by using only fifteen out of twenty amino acid composition features. The results showed that our RF-SBC-based method outperforms other existing algorithms in terms of accuracy, sensitivity, specificity and MCC measures in discriminating GPCRs from non-GPCR proteins.

Keywords:

GPCR , subtractive clustering , random forest , fuzzy logic , protein sequence