Re: Accounting for UCHIME chimeras when constructing OTU table

[Serena Thomson]

Hi there,

I have run pick_otus.py using UCHIME to highlight chimeric sequences
and would like to construct and OTU table. By using the seqs_otus.txt,
does this automatically have the chimeric sequences removed or do you
need to include the -e option (as in ChimeraSlayer)? In which case,
which file should you include for this when running the standard
UCHIME command in that both de novo and referenced based approaches
and union was adopted? There are two chimeric sequence files, do you
use both?

Serena

[Jose Carlos Clemente]

Hi Serena,

seqs_otus.txt already has removed the chimeric sequences, so you
should be good to construct the OTU table using it as an input.

Jose

[Serena Thomson]

Hi Jose,

Thanks for that. Just a more general question. I have processed the data a number of different ways ie chimera removed or not, de novo OTU clustering or referenced based etc and wanted to ask the following:

How is it that my chimera-removed data using UCHIME referencing against either UNITE or emerencia databases contain more total OTUs (although fewer total sequences) than the non-chimera removed, de novo clustered data? I would expect there to be fewer OTUs in the data where the chimeras have been removed. Is this not a correct assumption to make?

Many thanks

Serena

[Jose Carlos Clemente]

Hi Serena,

maybe there is a confusion here. In your original email (May 15) you said

I have run pick_otus.py using UCHIME...

does that mean you did OTU picking and chimera checking? So what you
have is result A (OTUs picked with usearch without chimera checking)
and result B (OTUs with usearch+chimera checking), but you get more
OTUs in B than in A, is this correct?

Jose

On Thu, May 17, 2012 at 8:07 AM, Serena Thomson

[Serena Thomson]

Hi Jose,

Sorry if I didn't make sense. I have run pick_otus.py specifying -m UCLUST (de novo) and no chimera removal step - (lets call this set A) and I have compared this with pick_otus.py specifying -m USEARCH (referenced against either UNITE or emerencia) which obviously includes a chimera removal step (but I have included the --cluster_size_filtering to avoid 'denoising' twice) - (set B). Set B however has more OTUs (but fewer total sequences) than set A. I would have expected that the data that has had chimeras removed would contain fewer OTUs...

Serena

[Jose Carlos Clemente]

Serena,

after some discussion with Tony, we think you probably cannot do a
comparison of the methods you are using directly. To compare, you
would need to run something like:

pick_otus.py -m usearch -x -k -l -i <input seqs> -o <no chimera
detection output> --word_length 64

which would disable the chimera check and size filtering, and

pick_outs.py -m usearch -l -i <input seqs> -o <chimera checking
performed> -f < reference database for chimera checking> --word_length
64

which does de novo an ref based chimera checking, with no size
filtering. However, the filtering of initial sequences during chimera
checking can change the seeds used (and numbers) for OTU picking in
the next step, so even in this case I'm dubious whether you'd be able
to compare results directly.

In any case, please remember we encourage our users to always check
sequences tagged as chimeric: most of them will be, but some of them
probably are not (and you will probably miss some that were never
tagged as such).

Jose

On Thu, May 17, 2012 at 9:53 AM, Serena Thomson