Comparisons of classification methods for viral genomes and protein families using alignment-free vectorization.

Abstract In this paper, we propose a statistical classification method based on discriminant analysis using the first and second moments of positions of each nucleotide of the genome sequences as features, and compare its performances with other classification methods as well as natural vector for comparative genomic analysis. We examine the normality of the proposed features. The statistical classification models used including linear discriminant analysis, quadratic discriminant analysis, diagonal linear discriminant analysis, k-nearest-neighbor classifier, logistic regression, support vector machines, and classification trees. All these classifiers are tested on a viral genome dataset and a protein dataset for predicting viral Baltimore labels, viral family labels, and protein family labels.