Error for specific dataset when using rand_formula "contrasts can be applied only to factors with 2 or more levels"

[tybonic]

Hi Frederick,

I am running ancombc2 on multiple subsets of my data containing different samples. I generate these subsets by filtering a phyloseq object containing all my samples. ancombc2 works fine with 3 of the 4 subsets, but for the 4th subset, I get an error message. This is my call;

ancombc2_out <-ancombc2(
    data = dat_phy_plant_2021_landrace,
    tax_level = NULL,
    fix_formula = "treatment",
    rand_formula = "(1 | run_nr_exp_013) + (1 | block)",
    n_cl = 28,
    verbose = TRUE
)

And this is the error I get:

Obtaining initial estimates ...

Error: Encountering the error for `lmerTest` package.
Please try to select one of your taxa and use its raw counts to fix the same linear mixed-effects model using `lmerTest` without the `ANCOMBC` package.
Load all necessary packages EXCEPT `ANCOMBC`, and see if the error arises due to package incompatibility or other issues.
The error message from `lmerTest` is as follows:
contrasts can be applied only to factors with 2 or more levels
Traceback:

1. ancombc2(data = dat_phy_plant_2021_landrace, tax_level = NULL, 
 .     fix_formula = "treatment", rand_formula = "(1 | run_nr_exp_013) + (1 | block)", 
 .     n_cl = 28, verbose = TRUE)
2. .iter_remle(x = x, y = y1, meta_data = meta_data, fix_formula = fix_formula, 
 .     rand_formula = rand_formula, lme_control = lme_control, theta = NULL, 
 .     tol = iter_control$tol, max_iter = iter_control$max_iter, 
 .     verbose = iter_control$verbose)
3. tryCatch({
 .     result <- lmerTest::lmer(formula = tformula, data = data.frame(y_crt = y[1, 
 .         ], meta_data), control = lme_control)
 . }, error = function(e) {
 .     message <- sprintf(paste("Encountering the error for `lmerTest` package.", 
 .         "Please try to select one of your taxa and use its raw counts to fix the same linear mixed-effects model using `lmerTest` without the `ANCOMBC` package.", 
 .         "Load all necessary packages EXCEPT `ANCOMBC`, and see if the error arises due to package incompatibility or other issues.", 
 .         "The error message from `lmerTest` is as follows:", e$message, 
 .         sep = "\n"))
 .     stop(message, call. = FALSE)
 . })
4. tryCatchList(expr, classes, parentenv, handlers)
5. tryCatchOne(expr, names, parentenv, handlers[[1L]])
6. value[[3L]](cond)
7. stop(message, call. = FALSE)

I tried fitting the model for one if my taxa with lmerTest:lmer() as recommended and that worked fine. Also, my phyloseq object has more than two levels for each random factor:

table(sample_data(dat_phy_plant_2021_landrace)$run_nr_exp_013)
table(sample_data(dat_phy_plant_2021_landrace)$block)

 exp_013_run1 exp_013_run2b 
           21            15 
1_2021_ru 2_2021_ru 3_2021_ru 
       12        11        13 

The factor variables have the data type "factor" and there are no NAs. I am a bit clueless on how to continue troubleshooting. Perhaps you have encountered this error before and have a hint on how to fix it?

[elmonten]

Hello!

I have been encountering the same issue. What transformation did you use?
I was using centered log ratio, but then I switched to microbiome::transform(pseq_genus, "rclr") and didn't show the error. Thus, I assumed it had something to do with it.