Beta group significance. High p-value in Permanova testing

Hi everyone,

I am trying to analyze microbiome content and obtain taxonomic heatmap and weighted UniFrac PCoA plot. I am trying moving pictures tutorial and each day something clarifies to later confuse me in following steps.
I have reached the last steps of tutorial - computed diversity metrix (core-metrix-results folder). I wanted to analyze sample composition in context of the groups I had in my metadata file using PERMANOVA test.

qiime diversity beta-group-significance
--i-distance-matrix core-metrics-results/weighted_unifrac_distance_matrix.qza
--m-metadata-file metadata-LB18_31.csv
--m-metadata-column Group
--o-visualization core-metrics-results/weighted_unifrac_distance_matrix.qzv

metadata-LB18_31.csv (5.2 KB)

weighted_unifrac_distance_matrix-pairwise.qzv (410.8 KB)
weighted_unifrac_emperor.qzv (781.4 KB)

I didn't understand why I got such high p-value. I decided to divide my sample and do the sample analysis for two groups divided into lifestyle 1 and 2 (as you can see in metadata file) but then p-values were even higher.

How can I fix this? What went wrong?

Thanks for your help!

Hey there @Jo_mee!

Probably because of the data, right?

You probably don't want to go down that road (e.g. "fixing" it) --- the signal is what the signal is. Nothing "went wrong," necessarily, it could just be that the results aren't significant. Have you had a chance to rerun this using the unweighted unifrac results? How about Bray Curtis? Do they show the same results? :microbe: :microscope:

1 Like

Hello @thermokarst,

Thanks for answer.
In the post I meant - can my qimme analysis pipeline lead me to such p-values?
It is clearly understandable it may be the matter of my results themselves.
However, I wanted to exclude qiime factor here.

I attach Bray Curtis and unweighted unifrac results. They seems slightly better but that is not what I expected to get.

unweighted_unifrac_distance_matrix.qzv (415.6 KB)

bray_curtis_distance_matrix.qzv (388.5 KB)

Yes, insignificant results are possible. These p-values are not unreasonable. In fact, some of the p-values are significant (p < 0.05).

If you suspect something is wrong, the best way to confirm would be to replicate these tests outside of QIIME 2, e.g., in R. You will need to carefully replicate the same filtering, etc, if you really do want a side-by-side comparison.

But this is much more likely to be a technical issue — just because you are not retrieving an expected result probably has more to do with how the samples are prepared and measurements are made. And how data are processed.

You are subsampling evenly at 1000 sequences. This could be problematic. Many other things could have gone wrong. This could be an issue with:

  1. you are subsampling too low. You need higher sequence coverage to recover useful signal.
  2. your denoising protocol is bad, e.g., you are trimming too much/too little and it is skewing your data.
  3. you could have contaminants or sequence errors of all sorts, adding noise to your data and drowning out the expected signal.

You should check for all of these possibilities, especially if, e.g., you see a significant effect with other statistical techniques but cannot replicate with this test.

But it is more likely that there is not a significant difference between your groups.

Why are you so surprised that you did not get a significant result? Have you tested with, e.g., QPCR or other techniques where you do see an effect that you just cannot replicate here?


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