I am running beta group significance to see what days are significantly different from other days, and I'm surprised by my results.
Based on the taxonomic bar plots of my samples from group 1 or group 2 (both over time as the x axis) if anything would be significant, I would expect it to be certain days in group 2 (for example day 1 and day 4). But when I run the analysis on each group separately, group 1 has days that show up as significantly different from one another, but not group 2. This is of course looking at the q values, after correcting for multiple comparisons.
Am I correct in that this is a surprising result, and suggests that something went wrong here? Am I making a critical error in how I am running my beta group significance command or is my understanding of this statistical test wrong?
Attached are the bar plots for the two different groups, as well as the results of the command. I tried other distance matrix's as well, with the similar results.
Group1-weighted-unifrac-DayPhys-significance.qzv (397.1 KB)
Group2-weighted-unifrac-DayPhys-significance.qzv (445.6 KB)
qiime diversity beta-group-significance
Any help would be appreciated,
thanks in advance for your time!
Your command looks reasonable, and makes sense for what you want to do. Given your group sizes, and number of comparisons you’re making, and the distribution of your group 1 boxplots, I’d say the statistical test is giving you an answer that makes sense. (I’m having some issues loading the boxplots for your group 2 visualisation.) You’re subject to a big variation in your data, based on the small sample size, which is a concern.
Without knowing the taxonomic level of your boxplots (phylum?), it can be hard to know for sure whether your intuition is good or not. For instance, if, in group 1, the yellow bar represents a large number of features with a large degree of evolutionary difference between them, it’s entirely plausable that your weighted UniFrac is different.
I have a different question, though. It looks like you’re handling time series data. If you’re doing repeated measures of the same thing, it might be possible to use the longitudinal plugin. The advantage of longitudinal analysis in microbiome data is that you remove some of inter-individual (inter environment) differences in the microbiome.
This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.