I have a question about get.communitytype (or what used to be called get.metacommunity).
As I understand it, this is used to see if community composition from different samples cluster together or not, and if so, what is the optimum number of clusters. So this can be used to determine for instance, gut enterotypes (albeit slightly different method to the one used by Arumugam et al). Is this correct?
Anyway, assuming this is correct, I ran the command, and on the screen (and in .fit file), K=3 is given as lowest Laplace value) which I guess means 3 clusters is the optimum for my data set (which is consistent with other reports). BUT when I look at all other files (.design, .parameters, etc), they partition all my samples into 1 (as opposite to 3).
Confused, I thought I was doing something wrong, so I ran the same command after converting all into phylotypes (at label=1) rather than in OTU, and this works; I get lowest Laplace value at K=3 and they partition my samples into 3 in all output files.
I performed many test in order to understand why it sometimes works, sometimes doesn’t.
And I found this: it depends on the number of processors… See below.
With processors=10, the K values are mixed up in the display (not the fit file) and only K=1 is retained, therefore all the samples are in partition 1.
Whereas without processors option, everything is fine… Apparently, there is a bug with this option.
