ANCOVA for ANCOMBC

General Discussion
1

Hey guys,

I need to perform ANCOVA such as the one performed at table 3 from this paper here (Gut Microbiota and Metabolome Alterations Associated with Parkinson’s Disease - PMC):

image
image1061×463 98.8 KB

I have seen that the q2-longitudinal plugin have ANOVA but I was wondering if by any chance there is a perspective to add ANCOVA. Anyway, if I understood correctly in this paper the ANCOVA was performed with the output from LEFse, and I intend to do it with ANCOM-BC output. Can anyone shed a light and help me understand if I should use the log fold change info, or the q-value? (Since I performed the ANCOM-BC from qiime2 with q-value, I believe I don't have to do the bonferroni correction).

Also, I am wondering if I can simply get tsv from ANCOM-BC and use it from all different variables without using a filter of p-value or q-value (since I have multiple variables - considered as covariates -
if I putted a filter of p-value < 0.05 for example it wouldn't be possible to analyze all variables at the same time because each variable returns a different output of taxa; the only way in my perspective is to analyze them all without any filter; I am putting an example of the tsv file of ANCOM-BC below)

tempsnip
tempsnip1724×448 20.3 KB

I am facing difficulties because I have never done this test before.

Thank you in advance.

1 Like
2

Good to see you again, Liv,

Let's start with your goal.

I need to perform ANCOVA such as the one performed at table 3 from this paper here (Gut Microbiota and Metabolome Alterations Associated with Parkinson’s Disease - PMC):

Okay. In this table, they identify possible microbial biomarkers.

If I were doing this analysis today, I would use ANCOM-BC instead of ANCOVA or LEFse because Microbiome Datasets Are Compositional: And This Is Not Optional

(If you are trying to replicate this table we could also help you do that!)

Both!
The log fold change is the effect size.
The q-value is the p-value adjusted for multiple-testing. This is a measure of significance.

(Since I performed the ANCOM-BC from qiime2 with q-value, I believe I don't have to do the bonferroni correction).

Correct. The q-value uses a Holm correction, which is "'uniformly' more powerful than the classic Bonferroni correction" anyway!

This is a modeling question. Unfortunately, I don't know enough about your experimental setup to comment on this. This question is best answered by a statistician on your team.

1 Like
3

Good to see you too, always helping me out :smile:

Ooh, possible microbial biomarkers? I didn't know that was possible. And also, is ANCOVA comparable to ANCOM-BC? Because it was asked me to do ANCOVA to adjust for covariables (like sex,age etc as reported in the original paper) to see how much they impact on the dependent variable, which I don't think is possible only with ANCOM-BC.

Yes, I am trying to replicate this table hehe I would be eternally thankful if you could help me with this task.

1 Like
4

Understood!

What is your dependent variable?

Looks like I messed up my link. Here's the article:

1 Like
5

The treatment (lithium treated and treated with other stabilizers).

6

Huh. I would expect that treatment (control vs lithium vs stabilizerA) would be the independent variable. Interesting study!

I think it's time to bring this to your PI or a stats person on your team. I don't know enough about your study design. Once you have a really solid understanding of the experimental design, you can start modeling it.

For stat testing and modeling I use R/Tidyverse and rstatix. They support ANCOVA!

2 Likes