The constantly decreasing cost and increasing output of current sequencing technologies enable large scale metagenomic studies of microbial communities from diverse habitats. Therefore, fast and accurate methods for taxonomic classification are needed, which can operate on increasingly larger datasets and reference databases. Recently, several fast metagenomic classifiers have been developed, which are based on comparison of genomic k-mers. However, nucleotide comparison using a fixed k-mer length often lacks the sensitivity to overcome the evolutionary distance between sampled species and genomes in the reference database. Here, we present the novel metagenome classifier Kaiju for fast assignment of reads to taxa. Kaiju finds maximum exact matches on the protein-level using the Borrows-Wheeler transform, and can optionally allow amino acid substitutions in the search using a greedy heuristic. We show in a genome exclusion study that Kaiju can classify more reads with higher sensitivity and similar precision compared to fast kmer based classifiers, especially in genera that are underrepresented in reference databases. We also demonstrate that Kaiju classifies more than twice as many reads in ten real metagenomes compared to programs based on genomic k-mers. Kaiju can process up to millions of reads per minute, and its memory footprint is below 6 GB of RAM, allowing the analysis on a standard PC. The program is available under the GPL3 license at:

Whole-genome sequencing is becoming the method of choice but provides redundant data for many tasks. ReadFilter (https://github.com/ssi-dk/serum_readfilter) is offered as a way to improve run time of these tasks by rapidly filtering reads against user-specified sequences in order to work with a small fraction of original reads while maintaining accuracy. This can noticeably reduce mapping time and substantially reduce de novo assembly time.