Prediction of human leukocyte antigen gene using κ-nearest neighbour classifier based on spectrum kernel

Human Leukocyte Antigen (HLA) plays an important role in the control of self-recognition including defence against microorganisms. The efficient performance of classifying HLA genes facilitates the understanding of the HLA and immune systems. Currently, the classification of HLA genes has been developed by using various computational methods based on codon and di-codon usages. Here, we directly classify the HLA genes by using the κ-nearest neighbour (κ-NN) classifier. To develop an efficient κ-NN classifier, we propose the use of a spectrum kernel to investigate HLA genes. Our approach achieves an accuracy as high as 99.4% of the HLA major classes prediction measured by ten-fold cross-validation. Moreover, we give a maximum accuracy of 99.4% in the HLA-I subclasses. These results show that our proposed method is relatively simple and can give higher accuracies than other sophisticated and conventional methods.