Item

Abstract

The most abundant forms of life on Earth, the microbes, dwell in human body and their roles in disease development have been studied. Infections by SARS, MERS coronavirus and more dreadful Ebola virus have shown high fatality rates and epidemic potentials since the outbreaks [1]. A fast isolation of virus- specific genomic contents from sputum or blood samples and an accurate detection of species within would enable a timely cope with the spreads of infections and further help to design vaccines. Metagenomics, a study on a mixture of microbiomes, has some primary questions: the quantitative estimation of species in a sample, and the isolation of target species in exclusion of background genomes. I introduce a highly sensitive and efficient taxonomy classifier, Poseidon, and present the results compared with recently developed algorithms, Kraken [2] and CLARK [3]. They associate a set of k-mers, sequences of length k, to the taxonomy identifiers. Poseidon builds a population index on top of the FM-index, a compressed self-index [4], where the genomes are stored in a compact form and classifies reads by backtracking with variable-length k-mers. The evaluation has been performed on simulated datasets of 8,294 species by Mason [5] and ART [6]. In all evaluations, Poseidon is the only algorithm keeping a high species-level sensitivity, and achieves the highest genus-level accuracy. In a clade exclusion experiment, the accuracy is measured for Proteus vulgaris while the database only comprises Proteus mirabilis, and Proteus penneri. Poseidon obtains Proteus-genus sensitivity of 41.1, which is around 16 higher than Kraken and CLARK with precision of 99.52. Poseidon is 9.4 times more memory-efficient than Kraken and 6.4 times than CLARK. The assignment speed is measured on 100-bp 6,929,444 reads. The speed of Poseidon and Kraken is similar while CLARK is 1.5 times faster than both algorithms. Poseidon will improve the quality of further analysis in microbiome-related researches through its high accuracy.