OTU overlap between samples?

[Diana]

HI All,

I'm looking for a way to ask if specific groups of samples are more
likely to contain the same OTUs than a randomized group of samples.
So, I'd like to be able to calculate the fraction of shared OTUs in a
group of samples (relative to the total number of OTUs in these
samples). And, I'd like to be able to permute a complete dataset to
ask if the observed number of shared OTUs is different from a random
expectation. Any idea if something like this is available in QIIME?

Thanks!!!
Diana

[Jens Reeder]

Hi Diana,

Qiime currently supports two analyses that might help you to achieve what you want:

1. otu_category_significance.py

(From the help:)
The script otu_category_significance.py tests whether any of the OTUs in an OTU table are significantly associated with a category in the category mapping file. This code uses, ANOVA, the G test of independence, or Pearson correlation to find OTUs whose members are differentially represented across experimental treatments or measured variables. It can also be used with presence/absence or abundance data for a phylogenetic group (such as that determined with quantitative PCR) to determine if any OTUs co-occur with a taxon of interest.

This test can be performed on a single OTU table or on a directory of OTU tables (for example, the output of multiple_rarefactions_even_depth.py). If the script is called on a directory, the resulting p-values are the average of the p-values observed when running a single test on each otu_table separately.


2. The svn version of Qiime has a basic scripts that computes the number of shared OTUs between all pairs of samples. Check out shared_phylotypes.py if you are interested.

One use case for this script would be the following:

a. Do multiple rarefaction with a certain number of sequences per sample using multiple_rarefactions_even_depth.py

b. Run shared_phylotypes.py on the output directory of step a.

c. Run dissimilarity_mtx_stats.py on the output_dir of step b.

In that way you get for each pair of sample the mean, and std dev and can compare if two sets of pairs are significantly different using a t-test outside of QIIME.

We do want to extend the shared OTU analysis in Qiime. Therefore, if neither of the above scripts do what you need, we would appreciate it if you can file a feature request at http://sourceforge.net/tracker/?group_id=272178&atid=1157167, briefly stating what the analysis should do and what kind of conclusion such an analysis would allow.

Jens