reference-base OTU picking and seed sequences

[pad]

Dear qiime,

I'm contemplating doing some form of ref-db OTU picking for Illumina
data but am not sure exactly how it works. While I understand that
sequences with hits (at whatever threshold) to a reference sequence
are retained and clustered, what is not clear to me is how uclust
assigns the seeds: does it just use the reference sequences as OTU
representatives or does it first determine whether the query reads
match the references at the given threshold and then cluster those
sequences that pass the threshold separately, i.e., using query
sequences as seeds? Would a ref-db approach be the preferred route for
Illumina data compared to a purely de novo clustering method? Because
I use a different barcode for each end, I am assuming that
concatenating the reads that strictly match the proper barcode
combination would be a reasonable way to deal with error. Your
thoughts?

Many thanks.
Pedro.

[Tony Walters]

Hello Pedro,

1.  If reference based OTU picking is used (ie, -m uclust_ref with pick_otus.py) the reference database sequences all act as seeds for the tested sequences to cluster against.  Depending upon whether new clusters are allowed, any sequences that fail to cluster at the given identity will be effectively added as new seeds to the existing set of seeds as OTU picking progresses.

2.  We do suggest doing reference based OTU picking for Illumina data, considering both the ability to parallelize the process and the tendency of short reads to fall into fewer clusters with the de novo process, which seems to be incorrect (the seed sequence is different than the seed sequence that would be used if the full 16S sequence was clustered at 97%).  If you want to do parallel OTU picking that allows new clusters, you can follow this tutorial here: http://www.qiime.org/tutorials/open_reference_illumina_processing.html

3.  Handling paired end data for OTU picking is tricky.  It will probably be easies to cluster just the forward or reverse reads independently and compare the results.

Hope this helps,

Tony Walters