How to overcome unexpected differences


I am running Qiime2 analysis on mice fecal samples from 2 similar experiments and looking for differences between treatment groups.
I got a significant microbial differences between my experiments that I didn’t expected to received due to the fact that the experiments were the same except for adding three treatment groups.
Is there a way to overcome these differences in order to search for differences between my treatment groups?
It should be mentioned that all samples were sequenced at the same time and the differences are also without the added treatment groups.

Thank you,

It’s always good to double check if there is no mistakes with a metadata file.
I recently got a new dataset and i found an errors in a metadata file I was provided and a lot of unexpected differences from the first analysis were perfectly explained after reruning the analysis with corrected metadata file.

1 Like

Hi @Dana,

This is a common problem in mouse experiments and entirely to be expected. Especially if they’re new batches of mice you ordered in from elsewhere. There’s a couple reviews on the issues, I happen to like this one by Nguyen et al. And so, while I also agree with @timanix, I think its probably a real issue and not one caused by mixed metadata.

As to what you can do with it… you may need to ackowledge that it exists and analyze seperately. Like, this is a driver of your microbial community and it sucks, but you can’t change it. You need to compare them against the control you included in each experiment, and then see if there’s additional replication that way. …Maybe show the two groups in one PCoA in hte supplement and then show them separately through out or something? (Paper writing strategies vary). There’s also questions about what your expectations are out of the experiment that frame how you want to analyze and present it, so think about those as well?



To add to @jwdebelius’s great answer, if you happen to have longitudinal data, or simply pre/post treatment then something you can do is calculate the differences in groups across time then compare those differences across groups, rather than compare groups directly to each other. That way you at least minimize the batch effect and account for starting microbiome differences.
On another note, did you analyse the 2 experiments separately? For ASV methods it is very important that you are comparing the exact size/location of sequences together, for example your trimming parameters during denoising (Dada2/Deblur) should be identical across runs.

1 Like

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