Alpha/Beta diversity plots and reporting specific samples/groups that show significant differences


I'd just like some advice regards interpreting my alpha and beta diversity analysis outputs. I have 8 compost samples (3 replicates each) chosen at different timepoints.

I have ran the Alpha diversity analysis for Shannon's diversity, Faith's phylogenetic diversity and Pielou's evenness and all report no significant differences amongst samples.

9_evenness_group_significance.qzv (437.7 KB)
9_faith_pd_group_significance.qzv (438.4 KB)
9_shannon_group_significance.qzv (437.7 KB)

Hypothetically speaking, if there was a significant difference reported, how could I determine which samples are responsible? Would it make sense to look at the reported pairwise comparisons and the associated p-values and corrected p-values (q)?

When I do my beta analysis (bray Curtis and unweighted unifrac) on the same set of samples they both report a significant p-value.

9_bray_curtis_significance_PW.qzv (564.3 KB)
9_unweighted_unifranc_group_significance_PW.qzv (575.7 KB)

Similar to the above is it possible to determine which specific sample comparisons are significantly different using these metrics.

Thanks in advance

HI @dfitzer1

Yes, looking at the pair-wise comparisons would give you the metrics to indicate samples responsible.

However, looking at your data the replicates may be problematic. For Kruskal-Wallis, and most of the statistical tests in Qiime2, one of the key assumptions is independence across samples. Replicates have inherent dependance. You will need to work around this by creating models that account for dependance.
This forum post regarding replicates should be helpful to you in treating your replicates.

The pairwise comparisons also will be helpful to you in this case for specific sample comparisons.

I hope this answer is helpful to you! Let me know if you have any other questions as you work through your analysis.

1 Like

Hello Hannah

Thanks for responding

I am not sure what you mean by problematic. Do you mean they are not independent as they are in triplicate? Would you suggest the 3 replicates for each timepoint be combined? I thought I had read elsewhere that this was not advised.

I could be misinterpreting this and your explanation.


Hi @dfitzer1,

Replicates can sometimes make statistical comparisons hard because we do not want to violate assumptions of independence. It's just good to be aware of that dependency as you continue!

Can you clarify your definition of replicates? Are these Technical Replicates (repeated measures of a sample) or Biological replicates (multiple samples from the same environment or subject)?



Hi @cherman2

Thanks for the response. They are Biological replicates.


1 Like

Hi @dfitzer1,

There shouldn't be any issues. I would just keep this dependency in mind as you are running analyses!


This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.