Hello, I have some questions regarding this topic, since I have occured the same problem in my data
experimental setup:
chicken experiment, chickens divided over 2 rooms, with 3 control groups and 3 experimental groups in each room.
My alpha diversity analysis (Faiths PD and pielou evenness) are for both room as well as treatment not significant.
my beta diversity analysis: I performed an adonis for room+treatment since I am interested in the effect that treatment has. I performed this adonis for both unweighted unifrac and weighted unifrac. weighted unifrac gave no significant p-value for treatment
unweighted unifrac gives a p value < 0.05 for treatment. So there are differences in beta diversity caused by treatment.
So I used ANCOM to look what these differences are... but I see no taxa in my ancom treatment. ancom-treatment.qzv (491.7 KB)
So my beta diversity indicates that there is a difference caused by treatment, but ANCOM cannot tell me what this difference is. Has this something to do with me only observing a significant difference for the unweighted unifrac and not the weighted unifrac?
No, its because those two statistical methods aren't consistent. There are no guarantees that beta diversity will agree with ANCOM. Such is the nature of the statistical methods ecosystem unfortunately.
There are new emerging tools that can directly address this issue like Birdman. Although its very much in early stages of development, so its not going to be easy to use...