A support vector machine based approach for Prediction and classification of polyanion Binding proteins

Abstract

Polyanionic proteins are extremely abundant in both extracellular and intracellular environment of the cell. These binding proteins contain multiple positively charged regions which are involved in phosphorylation processes that play a vital role in interaction with cellular proteins and also involved in process of protein folding. The various classes of polyanion binding proteins used are actin, heparin, heparin sulfate and tubulin binding proteins. In this article, we are interested in protein sequence based classification of polyanion-binding and non-polyanion binding proteins. Firstly, protein sequence features like amino acid composition, dipeptide composition, hydrophobicity and hybrid combinations of these features were used to develop SVM modules. Then training and testing cycle were performed using the SVM^light software. These modules were then evaluated using the 10 fold cross-validation technique. Furthermore, the method was able to predict major classes of binding proteins based on amino acid composition(AAC), dipeptide composition (DPC), hydrophobicity (Hydro), (AAC + DPC), (AAC + Hydro) and (DPC + Hydro) with an accuracy of 68.2012%, 70.2796%, 53.8530%, 76.6861%, 64.3378%, and 71.5340% and it was also able to predict major subclasses of polyanion binding proteins  using AAC, DPC , Hydro ,AAC + DPC,AAC + Hydro and DPC + Hydro with a maximum accuracy (92.80%, 94.44%, 93.30%, 93.29%, 93.40%, 93.33%),(57.69%, 57.94%, 57.32%, 74.67%, 59.73%, 57.16%),(84.61%, 88.88%, 85.78%, 86.67%, 86.67%, 86.67%) (80.76%, 83.22%, 82.61%, 82.31%, 77.73%, 78.44%)  for heparan sulfate, actin , heparin and tubulin respectively. We obtained a good classification performance for the SVM classifier trained with combined feature of amino acid and dipeptide features.