DNA Sequence Classification: It’s Easier Than You Think: An open-source k-mer based machine learning tool for fast and accurate classification of a variety of genomic datasets

Supervised classification of genomic sequences is a challenging, well-studied problem with a variety of important applications. We propose an open-source, supervised, alignment-free, highly general method for sequence classification that operates on k-mer proportions of DNA sequences. This method was implemented in a fully standalone general-purpose software package called Kameris, publicly available under a permissive open-source license. Compared to competing software, ours provides key advantages in terms of data security and privacy, transparency, and reproducibility. We perform a detailed study of its accuracy and performance on a wide variety of classification tasks, including virus subtyping, taxonomic classification, and human haplogroup assignment. We demonstrate the success of our method on whole mitochondrial, nuclear, plastid, plasmid, and viral genomes, as well as randomly sampled eukaryote genomes and transcriptomes. Further, we perform head-to-head evaluations on the tasks of HIV-1 virus subtyping and bacterial taxonomic classification with a number of competing state-of-the-art software solutions, and show that we match or exceed all other tested software in terms of accuracy and speed