confused about the newest version result

[Big]

Dear Nicola, I tried the new version by shuffling 2 paired end fastQ files into one, but the result shows it contains 68% cyanobacteria, which is hugely different from the older version on the same sample: no such bacteria should be seen in vaginal samples. Any possible reasons?

Thanks!
B

[Nicola Segata]

Hi Bing,

 what kind of filtering did you apply to produce the fasta file you used originally? When using fastq with the new version, MetaPhlAn will take into account the quality of the reads but of course it doesn't apply any specific filtering (like removing possible contamination, or human reads, or setting a threshold on the read length).

Any chance I can have a look to the sample for a more specific answer? In case let me know how we can arrange the transfer (with a private email). Or is the sample contained in the dataset we already have on our server?

thanks

Nicola

[Nicola Segata]

I'm briefly reporting the conclusions of the above issue. 

It seems that BowTie2 is sometimes overly sensitive when applied to non-preprocessed fastq files when using a local (rather than global) hit acceptance policy. The issue is not occurring when requiring global alignments in BowTie2.

The recommendation (posted also on the MetaPhlAn webpages) is thus to use the "--bt2_ps very-sensitive" or "--bt2_ps sensitive" options instead of the default "--bt2_ps very-sensitive-local" when using MetaPhlAn with BowTie2 and fastq input files.

many thanks

Nicola