Compressive sampling for phenotype classification

Genome classification has become an increasingly important genomic research method for cancer identification and treatment. One challenge associated with genome classification is feature selection; which genes can be used for phenotyping. This challenge is made more complicated considering affected gene mutate at different rates and schedules. In addition, the number of genes and consequently the number of base pairs cause genome classification to suffer from the curse of dimensionality. Some previous classification efforts rely on variant detection for feature selection. Variant detection aligns a target genome to a reference genome in order to identify the variations between the two. Other methods are sequence-based, which involve generating features from base pairs of genome sequences. This paper details a sequence-based methodology developed using k-length subsets of the genome sequence, k-mers, as features for phenotype classification. As a result of the methodology introduced here, we remedied challenges associated with high-dimensional data and improved phenotype classification performance using sequence-based classification.