Classification of cereal proteins related to abiotic stress based on their physicochemical properties using support vector machine

Abiotic stress factors severely limit plant growth and development as well as crop yield. There is a great need to develop understanding of plant physiological responses to abiotic stresses in order to improve crop productivity through crop improvement programmes. Proteins play a central role in plant adaptations under stress and hence their identification is important to the biologist. Identification of such proteins by wet lab experimentation is sometimes expensive and timeconsuming. In such a situation, in silico approaches can be used to narrow down this search. In this study, classification of cereal proteins subjected to four different stresses, namely, extreme temperature, drought, salt and abscisic acid (ABA) was undertaken. Classification models were built using support vector machine (SVM) to predict the function of proteins under these abiotic stresses on the basis of 34 physicochemical features extracted from the protein sequence. Specific features of the protein sequence that are highly correlated with certain protein functions were selected by stepwise logistic regression, a feature selection method. SVM was trained using different kernel functions and cross-validated using 10-fold crossvalidation technique. Prediction precision was assessed through different measures such as sensitivity, specificity and accuracy. The accuracy of protein function prediction using SVM with different kernel functions ranges from 60% to 100%.