Issues with beta diversity comparison

[Kate Blackwell]

• Here's what I am trying to do:
  
• Compare the differences resulting from running different permutations of a bioinformatics pipeline on one samples using beta diversity (weighted unifrac) with Procrustes 
• Sample A w/permutation:
1.  Chimera Method: USearch61 de novo, OTU Picking Method: USearch61, Taxonomic Assignment Method: Blast, Singletons Included: No

2. Chimera Method: USearch61 de novo, OTU Picking Method: USearch61, Taxonomic Assignment Method: Blast, Singletons Included: Yes

3. Chimera Method: USearch61 de novo, OTU Picking Method: USearch61, Taxonomic Assignment Method: UClust, Singletons Included: Yes 

4. Chimera Method: USearch61 de novo, OTU Picking Method: UClust, Taxonomic Assignment Method: Blast, Singletons Included: No

5. Chimera Method: USearch61 de novo, OTU Picking Method: USearch61, Taxonomic Assignment Method: UClust, Singletons Included: No

6.  

   Chimera Method: USearch61 de novo, OTU Picking Method: UClust, Taxonomic Assignment Method: UClust, Singletons Included: Yes

7.  

   Chimera Method: USearch61 de novo, OTU Picking Method: UClust, Taxonomic Assignment Method: UClust, Singletons Included: No

• Having trouble with:
• Generating beta diversity measurements

When I run beta_diversity.py on the OTU table from a single permutation, it results in measured values of 0 for weighted unifrac as there is only one sample for each permutation.  

I don't think I can use beta_diversity.py on the different OTU tables from all of the permutations in one command line as each permutation has a different -t tree path and QIIME won't let me use a directory like it would for the OTU tables.

Do I need to go back and redo my analysis with more than one sample for each of the permutations in order to be able to compare the weighted unifrac values between samples?  Or is there a way to generate weighted unifrac values when comparing between the permutations?  I thought about merging the OTU tables into a single one, but I don't think that is an option as I would still have multiple -t tre paths and I used the pick_open_reference_otus.py command, which includes a de novo step and I have different #s of OTUs for each permutation.

Any suggestions or help is greatly appreciated!  Let me know if additional details are required.

Also, if you can think of an alternative analysis beside comparing the beta diversity analyses between permutations, I am open to suggestions.  I have the alpha diversity values too and was thinking of testing to see if they were sig. different from one another.  I was also comparing the OTU table results in a bar chart of just the major classes (anything greater than 1% relative abundance) to see if the phylotypes differed visually between permutations.

[jonsan]

Hi Kate,

Glad to see you're doing a sensitivity analysis! I've found this sort of thing to be extremely informative with my own data. 

A few relevant issues:

1) Beta diversity is a measure of the diversity differences *between* samples, while alpha diversity is the difference *within* samples. That's why you have zeroes for your bdiv output when analyzing a single sample -- between-sample differences require there to be more than one sample. 

2) Similarly, Procrustes analyses are a means to compare points in multidimensional space. For PCoAs (like those calculated based on beta-diversity distances), the number of dimensions in the space is equal to the number of the samples being compared minus 1. You can easily visualize this with two- and three-sample datasets. When comparing the distance between two samples (two points), you just need one dimensions -- a line. When comparing three samples, you just need a cartesian grid. 

3) Unifrac distances do not depend on taxonomic assignments, just on the OTUs generated. Taxonomy assignment happens after OTU have already been picked. Thus from the standpoint of your described analysis (Procrustes comparison of Unifrac distances), your parameter combos 1&5, 2&3, and 4&7 are identical -- they should lead to the same OTU clusters (not counting the potential for random variation in the OTU picking algorithm) and hence the same beta diversity measurements.

Hope that helps,

-jon

[Kate Blackwell]

Thanks, you confirmed my thoughts with this!  The major limitation I am facing with the most recent runs for comparison purposes is only doing it with one sample.