Simulated data for one cell type

[ahassien]

Hi there,

The data that I am working with has a single cell type (tumor cells), and I am wondering why this would not work within the scDesign3 framework. I am trying to benchmark the identification of DEGs.

I tried replicating the issue by changing the cell type filtering in the DE analysis example to only include one cell type: https://songdongyuan1994.github.io/scDesign3/docs/articles/scDesign3-DEanalysis-vignette.html#identification-of-de-genes-between-discrete-cell-types

# selected_cells <- which(colData(example_sce)$phenoid %in% c("b.cells","regulatory.t"))
selected_cells <- which(colData(example_sce)$phenoid %in% c("b.cells"))

And it yields the same error that I get when I run it on my own data:

Convert Residuals to Multivariate Gaussian
Error in `colnames<-`(`*tmp*`, value = rownames(sce)) : 
  attempt to set 'colnames' on an object with less than two dimensions
In addition: Warning message:
In mclapply(seq_len(n), do_one, mc.preschedule = mc.preschedule,  :
  all scheduled cores encountered errors in user code

Is there any reason why the scDesign3 framework requires more than one cell type?
Unrelated to the example, I am having some follow-up issues with my particular dataset, particularly involving the following error:

Error in x$fit : $ operator is invalid for atomic vectors

I am thinking that this error could be related to the one cell type issue since I do not want my formula to include cell_type.
Many thanks in advance!

[dsong-lab]

@ahassien Hi Alexa,
Yes, your guess is correct. Because here you only have one cell type, you should change any formula (e.g., mu_formula = 'cell_type') to mu_formula = '1'. You can imagine that if your covariate only has one value, it equals to the intercept. Let me know if you have more questions.

Best regards,
Dongyuan

[ahassien]

@dsong-lab Hi there Dongyuan,

This makes perfect sense. Thank you for the clarification. Does this also affect models with additional covariates?
For example, if I want to include 'sample' and 'grade' covariates in my mu formula. I get the following error regardless of whether I use mu_formula = "sample + grade" or mu_formula = "1 + sample + grade"

Input Data Construction Start
Input Data Construction End
Start Marginal Fitting
Marginal Fitting End
Start Copula Fitting
Error in x$fit : $ operator is invalid for atomic vectors

Here is my code for using the scDesign function:

simu_res <- scdesign3(sce = sce_test, 
                      assay_use = "counts", 
                      celltype = "cell_type", 
                      pseudotime = NULL, 
                      spatial = NULL, 
                      other_covariates = NULL, 
                      mu_formula = "sample + grade", # alternatively I have tried "1 + sample + grade"
                      sigma_formula = "1", 
                      family_use = "nb", 
                      usebam = FALSE, 
                      corr_formula = "1", 
                      copula = "gaussian", 
                      DT = TRUE, 
                      pseudo_obs = FALSE, 
                      return_model = FALSE)

Many thanks in advance,
Lexie

[dsong-lab]

Hi Lexie,
You need to add those covariate names into other_covariates; otherwise, scDesign3 does not know you have them in your dataset.

Best,
Dongyuan