Hi,
I am analyzing a dataset where there are 4 different treatments to mice that are housed by 2 or 3 per cage. (The group sizes are 5, 6, 9, 9).
I have run the adonis on qiime:
qiime diversity adonis
--i-distance-matrix bray_curtis_distance_matrix.qza
--m-metadata-file ../../metadata_CC.csv
--p-formula "group+cage"
--o-visualization adonis_bray_group_cage.qzv
I get this table:
|
Df |
SumsOfSqs |
MeanSqs |
F.Model |
R2 |
Pr(>F) |
| group |
3 |
1.960725 |
0.653575 |
4.090156 |
0.232278 |
0.001 |
| cage |
9 |
3.923902 |
0.435989 |
2.728475 |
0.464846 |
0.001 |
| Residuals |
16 |
2.556676 |
0.159792 |
NaN |
0.302877 |
NaN |
| Total |
28 |
8.441303 |
NaN |
NaN |
1.000000 |
NaN |
The cage seems to have a bigger effect on the data than the treatment group. If I switch the order of cage and group in formula, then the row for group disappears from output.
In the end I would like to know if there is still any differences between the groups after taking into account the cage effect. Is adonis the best way to tackle it? The discussions found on internet on how to eliminate batch effect from microbiome data seemed not very conclusive...
Any advice is highly welcome!
Thank you!
Hello Rahel,
Thank you for telling me about your experiment. I think I know what's going on here.
If I switch the order of cage and group in formula, then the row for group disappears from output.
The issue here is that your ~cage variable is nested within ~group, i.e. each cage comes from one group so no comparison can be made.
Cage nested within group:
Cage 1: Group A
Cage 2: Group A
Cage 3: Group B
Cage 4: Group B
Building and Group fully blocked:
Building 1: Group A
Building 2: Group A
Building 1: Group B
Building 2: Group B
Because this blocking is baked into your study design, there's no statistical trick you can do to change it.
The results you show come from the formula ~group+cage, and that seems OK to me! You are testing for group, and it shows a large effect (23%!!) and is significant (0.001).
The built environment often has a strong effect on mouse models. See this paper I worked on in 2016: Influence of early life exposure, host genetics and diet on the mouse gut microbiome and metabolome | Nature Microbiology
2 Likes
Hi Colin,
Thank you for your reply and explanation. It's reassuring to know that the formula ~group+cage is ok.
And indeed the cage effect has a major influence on the mouse microbiome, but I guess there's not much I can do at this point to take into account that effect.
Thanks!
1 Like