Applying get.communitytype to OTUs vs. phylotypes


I am interested in applying the Dirichlet multinomial mixtures (DMM) model proposed by Holmes et al. (2012, PLoS ONE 7(2): e30126) and implemented in Mothur via the get.communitytype command for an analysis I am currently working on. In the example provided by Holmes, and in Pat’s recent publication in Nature (2014 May 15;509(7500):357-60), the DMM model was applied to genus-level phylotypes, as opposed to an OTU table. In several other examples I came across (La Rosa PS et. al., 2012. PLoS ONE 7(12): e52078.), the DMM model was also applied to sequences clustered at a single taxonomic level.

I am hoping to get some feedback on what are the benefits and limitations of applying the DMM model to a table of genus-level (or other) phylotypes vs. an OTU table? Basically, I know I can fit the DMM model to an OTU table, but I am wondering if are there any good reasons as to why I shouldn’t? In particular I am thinking of tables of counts (as opposed to presence/absence).

Also, for the phylotype analyses, it seems some have chosen to “merge” sequences reads that cannot be classified at a given level (i.e. genus-level) to a single group denoted as “other”, while some have culled them from the table (if I am reading this correctly).

Thanks in advance for any thoughts on the topic!

I’ve found that at a finer taxonomic scale (OTU), there distribution of the OTUs’ abundances are just too flat to get good clusters. If you had more samples or there is less heterogeneity in your data, you could probably do OTUs. This was why I ultimately went with the genus-level approach in the Ding & Schloss paper.



Thanks so much for the thoughts and for providing your rational for your choice in the paper. It is much appreciated.