Hello. I am using metagenomeSeq for a metagenomic dataset.

I performed two models using fitZig.

Model I, qualitative analysis using two groups (e.g., obesity status: control vs. obesity) as independence variables.

Model II, quantitative analysis using a continuous variable (e.g, BMI) as independence variables.

Additionally, I used some covariates (e.g, age, sex..) in both two models.

I have some questions in the results of MRfulltable after the analyses.

Question 1.

What is the p-value? Is it the model's p-value or the BMI's p-value?

I don't know whoes it is.

I used 5 variables including 1 independence variable and 4 covariates.

In MRfulltable, I found 5 coefficients, one pvalue, and one adjPvalue.

Is the pvalue the first independence's (obesity status, my interested variable)?

If so, I'm wondering how the fitZig can recognize my interested variable among 5 variables.

Does the fitZig recognize the order of independent variables?

I found the different results when I changed the order of independent variables.

Isn't there p-values about covariate?

If I want to know the p-values of covariates (sex, age...), how can I know them?

Question 2.

I perfomed Model II after class-conversion of BMI using as.numeric(BMI).

I found differences in result table comparing to qualitative analysis (Model I).

The "+samples in group 0" was filled with "0" and "+samples in group 1" was filled with from "0" to "5".

What does it mean?

I don't know why there are groups although Model II is quantitative analysis.

Question 3.

How can I interpret the coefficient?

Is it 2log fold change?

Is it beta coefficient?

I'm sorry but statistics is too difficult for me. T.T

Question 4.

How is the oddsRatio defference from the coefficient?

It's difficult to interpret the MRfulltables.

Question 5.

In Model II, what does the oddsratios mean?

As you know, the continuous variables (BMI) is not categorical group.

Question 6. (Final)

I cannot clearly understand fitZig model yet.

I know that zero-inflated model consists of count-part analysis (only positive counts) and zero-part analysis (zero vs. non-zero).

Which is fitZig, count-part or zero-part?

Sorry, that's a lot of questions.

Many thanks in advance!

Thanks for your kindly answers.

I tried to use the "by" option but it's not woking with error messages except by=2.

I want to know the significance of effects of continuous variables as independent variables in fitzig model.

Q6. Is the fitzig possible for only categorical comparison?

Q7. I'm not sure whether my model was adjusted with controlling for covariates.

My all variables consist of quantitative variables including covariates in Model II.

The covariates start from 3rd column.

When I try to use the "by=3" , the follows error appear:

>head(MRfulltable(fit, by=3))

During startup - Warning messages:

1: Setting LC_COLLATE failed, using "C"

2: Setting LC_TIME failed, using "C"

3: Setting LC_MESSAGES failed, using "C"

4: Setting LC_MONETARY failed, using "C"

Error in head(MRfulltable(fit, by = 3)) :

error in evaluating the argument 'x' in selecting a method for function 'head': Error in rowSums(yy[, groups == 1]) :

'x' must be an array of at least two dimensions

I edited the post above additionally to include this message:

Additionally, if you have a non-categorical coefficient you should not be using MRfulltable. That function in particular runs fitPA on your (assumed) twp groups of interest. See ?MRtable, ?MRcoefs for alternative table outputs. MRcoefs will not produce the same error since it does not report the number of samples in either group 1, group2 (hence ideal for your continuous variable situation)

Thanks Joseph for your help with all of my questions.

As your suggestions, I tried to use the MRcoefs for a non-categorical variable with changing counts of the "by".

It worked fine without any error. I was able to find both coefficients and p-values of my covariates.

I think that my problem was non-categorical variables.

Thank you very much again, Joseph.