Maple will not work in our benchmarking setting

[Jieran-S]

Hi,

Maple is a model that requires Multi-sample dataset input to generate labels for individual samples (see publication (esp Fig. 1) and the code snippet below), whereas in our cases we input individual samples from a dataset. This method might not work under our framework.

The code from their fit_maple function:

  if(is.null(covars))
  {
    print("Note, no supplied covariates. Will use orig.ident as sample indicator.")
    sample <- as.factor(meta[,"orig.ident"])
    W <- as.matrix(stats::model.matrix(~ sample))
    rownames(W) <- rownames(meta)

Whenever I run it within our system, it returns an error as generally our orig.ident has only one level for each sample we input.

error message:

Error in h(simpleError(msg, call)) : 
  error in evaluating the argument 'x' in selecting a method for function 'as.matrix': contrasts can be applied only to factors with 2 or more levels

@shdam Could you check if this is the case or any workaround? I can put Maple as a low-priority implementation for now.

[shdam]

Hmm, I thought it would work anyway since the single-sample tests went well.. Oh well, it is built on spruce, which may then be the actual method that should be implemented, as that is for single samples.

[shdam]

I just saw that spruce is no longer maintained, so perhaps we should exclude maple as well, as it uses spruce.

[Jieran-S]

If it can work without error, I'd be happy to include the method. But I keep on getting the aforementioned error message. Did you have the same experience? Do you think there's any workaround?

[shdam]

Hmm, it would probably need some universal covariate to not use sample ID. However, using spruce instead would require minimal changes. Essentially only these two lines would need changing: library(maple) -> library(spruce) - since it is already installed in the environment. And: maple_results <- fit_maple( -> maple_results <- fit_spruce(

Of course, it would then make sense to rename the method to spruce, but this change should make the method run, and maple is essentially just a multi-sample version of spruce anyway.

[Jieran-S]

Aha I see! Thanks for the suggestion. I will try to run it after the modification.

We should also dicuss a bit whether to keep the method in the next meeting, given that it's no longer maintained.

[shdam]

Yeah, it's unfortunate since it seemed like a decent method, but the P.I. behind it now works in industry. I am curious to hear how it performs :)

[Jieran-S]

Changed to spruce and it seems to work. Close the issue for now