CLARK: fast and accurate classification of metagenomic and genomic sequences using discriminative k-mers

Abstract

BackgroundThe problem of supervised DNA sequence classification arises in several fields of computational molecular biology. Although this problem has been extensively studied, it is still computationally challenging due to size of the datasets that modern sequencing technologies can produce.ResultsWe introduce Clark a novel approach to classify metagenomic reads at the species or genus level with high accuracy and high speed. Extensive experimental results on various metagenomic samples show that the classification accuracy of Clark is better or comparable to the best state-of-the-art tools and it is significantly faster than any of its competitors. In its fastest single-threaded mode Clark classifies, with high accuracy, about 32 million metagenomic short reads per minute. Clark can also classify BAC clones or transcripts to chromosome arms and centromeric regions.ConclusionsClark is a versatile, fast and accurate sequence classification method, especially useful for metagenomics and genomics applications. It is freely available at http://clark.cs.ucr.edu/.Electronic supplementary materialThe online version of this article (doi:10.1186/s12864-015-1419-2) contains supplementary material, which is available to authorized users.

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

“…For contiguous k-mers, the classification precision increases as we increase k. However, the highest sensitivity occurs with somewhat shorter k-mers. Clark is more precise for long contiguous k-mers (e.g., k = 31), but the highest sensitivity occurs for k-mers of length 19-22 [19]. As a consequence, we considered here spaced seeds of length k = 31 and weight w = 22.…”

“…The classification algorithm of the "Spaced" mode is identical to that of the "full" mode (extensively described in [19]), except for two differences, namely (i) Clark-S queries against discriminative spaced k-mers instead of discriminative k-mers and (ii) Clark-S does three queries for each k-mer in a read, because there are three different databases. Finally, as done in the full and other modes, the read is assigned to the target that has the highest amount of successful queries, and several statistics (such as the confidence score and gamma score, see [19]) are computed as well.…”

“…. , p}, the set D i of discriminative k-mers for target g i is the set of all k-mers in g i that do not appear in any other reference sequences [19]. Given a spaced seed s of length k and weight w, we define D i,s to be the set of all w-mers obtained via s from k-mers in D i .…”

“…We have recently proposed Clark in [19] and demonstrated that its classification speed is currently unmatched.…”

“…Classification tools such as NBC [21], Kraken [25], Clark [19], among others, can be used to determine the composition of the microbial diversity from the sequenced reads for a microbial sample. We have recently proposed Clark in [19] and demonstrated that its classification speed is currently unmatched.…”