KrakenUniq+Bracken

[YiJessePi]

Hello, We are currently using kraken2+bracken but would definitely want to use the advantages of krakenuniq to identify unique kmers to reduce false classification. I came across this publication who states:

"If counting all listed taxa, regardless of low readcounts, the Kraken2 is known to produce many false positives [32], i.e., list taxa as present when they are in fact not. The KrakenUniq software has been developed to handle this problem [32]. We ran it to classify the metagenome reads for both healthy and IBD metagenomes. The overall results from both Kraken2 and KrakenUniq tools were similar, but KrakenUniq also reports the number of unique k-mers in each genome covered by the reads. On the other hand, only Kraken2 reports are compatible for running the Bracken software. Since we were interested in both—that is finding the true positive identifications, and their estimated abundances—we combined the two approaches."

Is krakenuniq indeed not compatible with bracken? What do you think about this suggested flow? kraken2->filtering by krakenuniq->bracken?

[salzberg]

Jesse, please email me ([email protected]) and I'll reply with more details that way. But that paper is not correct, Bracken is compatible with KrakenUniq, and we have used it that way many times. I can put you in touch with Jen Lu (who wrote the Bracken code) if you have issues. Your suggested flow might be a good one if you need to use kraken2 for the lower memory footprint, and then follow it with krakenuniq to eliminate false positives. Or you can just run krakenuniq->bracken. The latest kraken2 has a lower computational false positive rate, but it is not zero. With krakenuniq it is zero; i.e., if a hit is reported, you can be 100% certain that the read shares at least 1 k-mer with the genome.