Higher Classification Accuracy of Short Metagenomic Reads by Discriminative Spaced k-mers

The growing number of metagenomic studies in medicine and environmental sciences is creating new computational demands in the analysis of these very large datasets. We have recently proposed a timeefficient algorithm called Clark that can accurately classify metagenomic sequences against a set of reference genomes. The competitive advantage of Clark depends on the use of discriminative contiguous kmers. In default mode, Clark’s speed is currently unmatched and its precision is comparable to the state-of-the-art, however, its sensitivity still does not match the level of the most sensitive (but slowest) metagenomic classifier. In this paper, we introduce an algorithmic improvement that allows Clark’s classification sensitivity to match the best metagenomic classifier, without a significant loss of speed or precision compared to the original version. Finally, on real metagenomes, Clark can assign with high accuracy a much higher proportion of short reads than its closest competitor. The improved version of Clark, based on discriminative spaced k-mers, is freely available at http://clark.cs.ucr.edu/Spaced/.