weighted_unifrac_emperor removed some samples!

Hi friends,
I have 12 groups (with three samples each) = 36 samples! I wondered when I visualized my weighted_unifrac_emperor. I should have 12 orange, 12 blue and 12 green points on the photo, but foe example there are just 9 orange points! It means three orange point or samples are missing! I saw it in different angles but I did not see the others. I want to know what happened to them? By the way, what are the numbers above each axis?

Thanks
Qiimer

Hello @TurboQiimer,
My first thought is that they sampling depth was too low for the sampling depth you set at the Qiime core-metrics step :qiime2: . If that is the case they wouldn't show up in most of the diversity metric figures.
This is my first thought as I have had this happen to me. If those points are really important to the analysis then you might need to lower your sampling depth so they are included. However, This is just a guess. I would need more information on your pipeline in order to make a better suggestion.

Here is what I would do as a course of action, 1) check to make sure it is not the sampling depth. if it is great! if not 2) Upload some visualizations like your table.qzv and we will try to debug more!

As for the numbers at the top of the axis, that is the percent variation explained by that one dimensional axis.

Hope this helps!
Chloe :turtle:

2 Likes

You hit the nail on the head!
Yes, you are right! The sampling depth is too low! but I never thought that lower depth makes such a result.
So another question: Does low sampling depth make my analysis problematic? can I rely on the result? The result makes very sense with my gases results! But losing some point concerned me.
By the way, could you please give me more info/details regarding axis percentages? How they are produced?
Thanks a lot Dear @cherman2
Qiimer

Hello @TurboQiimer,
The long and short of sampling depth is that it is a balancing game.
If your sampling depth is lower, you get more of your samples in your data but you are not investigating each read at a “deeper” depth so you will lose features. If your sampling depth is higher, you will lose samples but be able to investigate the sample you have with a deeper sampling depth.

So it just boils down to what is best for your study. I mentioned that I ran into this problem as well. My solution was to lower my sampling depth because I had a small data set and it was important to have those samples. Maybe to you it is more worth it to throw out the some of your data points to be able to run your analysis with a deeper depth. At this point it is really up to you and the goals of the analysis.

As for the axis percentages, here is an article that talks about use of UniFrac in the microbial community, that might scratch the itch of what UniFrac is really telling you. Here is also a discussion that might help!

Chloe :turtle:

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