I recently used the minION (Nanopore, 9.4 flow cell, RAD001 kit) to generate a metagenome out of environmental samples. Passed reads weren't brilliant (196, average 1,594bp lenght), but working with centrifuge the classification outputs turned out to have quite low hitLength to queryLength scores (average 2%, max 14%). This plus score values don't give me a lot of confidence towards the results I got.

Has anyone else used centrifuge and experienced the same?

  • 3
    $\begingroup$ Sorry, but "has anyone else seen this" isn't on topic here (that's just a discussion and asking for personal opinion). Could you edit your question to make it request a solution to a specific problem instead? Something like "what sort of threshold should I use for centrifuge" or similar? $\endgroup$
    – terdon
    May 31 '17 at 13:03
  • 1
    $\begingroup$ Yes, this question can be improved, but I think people can understand what OP really wants. I don't think it deserves a closing. $\endgroup$
    – user172818
    Jun 1 '17 at 15:31

You should also keep in mind that single read accuracy of ONT still is a bit lacking, a 2D or 1D^2 accuracy of 95% still means that on average there's one error every 20 bp, and due to the nature of the data some stretches may be more junky than others. Maybe centrifuge doesn't like that.

Getting a feel of what the data can tell you (and what not) is pretty important.

You may want to try other classification tools to get an idea how well centrifuge works, Kraken comes to mind. But also: what about BLASTing some of these reads at NCBI and simply look at the results to get a first impression?


Not the answer you're looking for? Browse other questions tagged or ask your own question.