Error: length of 'dimnames' [2] not equal to array extent

[hl-xue]

Hello,

I encountered a problem running SampleQC with calc_pairwise_mmds:

sample-level parts of SampleQC:
  calculating 3 sample-sample MMDs:
  |==========================================================================================================| 100%

  clustering samples using MMD values
  calculating MDS embedding
Error in dimnames(x) <- dn : 
  length of 'dimnames' [2] not equal to array extent
In addition: Warning message:
In cmdscale(mmd_mat, k) : only 1 of the first 2 eigenvalues are > 0

Here is my script.

qc_dt <- make_qc_dt(merged_obj@meta.data, sample_var = "sample_id", 
                    qc_names = c("log_counts", "log_feats", "logit_mito"),
                    annot_vars = "condition") # Till here is fine
                    
qc_obj <- calc_pairwise_mmds(qc_dt, qc_names = c("log_counts", "log_feats", "logit_mito"),
                             annots_disc = c("cell_id", "sample_id", "condition"),
                             annots_cont = NULL) # The error occurs

Do you have any idea about this please? Thanks!

[alanaleah89]

Same Issue

qc_dt = make_qc_dt(seu.obj@meta.data, 
  sample_var  = 'RunID', 
  qc_names    = c('log_counts','log_feats','logit_mito','logit_rb','splice_ratio'),
  annot_vars  = NULL
  )
qc_names    = c('log_counts','log_feats','logit_mito','logit_rb','splice_ratio')
annots_disc = c('SampleName')
annots_cont = NULL
qc_obj = calc_pairwise_mmds(qc_dt, qc_names, annots_disc = NULL, annots_cont = annots_cont, n_cores = 5, one_group_only = TRUE)

Output

sample-level parts of SampleQC:
  calculating 6 sample-sample MMDs:
  |======================================================================================================| 100%Warning: 'package:stats' may not be available when loadingWarning: 'package:stats' may not be available when loadingWarning: 'package:stats' may not be available when loadingWarning: 'package:stats' may not be available when loadingWarning: 'package:stats' may not be available when loading

  clustering samples using MMD values
  calculating MDS embedding
  calculating UMAP embedding
Found more than one class "dist" in cache; using the first, from namespace 'spam'
Also defined by ‘BiocGenerics’
Found more than one class "dist" in cache; using the first, from namespace 'spam'
Also defined by ‘BiocGenerics’
Error in uwot(X = X, n_neighbors = n_neighbors, n_components = n_components,  : 
  n_neighbors must be smaller than the dataset size

[wmacnair]

Hi @hl-xue, @alanaleah89

Sorry I've been slow to respond, tricky with a job now... 😅

You have two slightly different issues, driven by the same problem: I developed SampleQC to deal with a problem I had with doing QC for a massive dataset with 150+ samples, and quite a lot of variability between those samples. This means that the code is not super well tested for small numbers of samples.

@hl-xue, in your example, what SampleQC is doing is calculating a type of pairwise distance between all pairs of samples, which it will then use to group the samples, and also find a way of plotting which samples are similar / different. This makes sense when you have a large number of samples, but you only have 3! This is the error - it tries to do a PCA-like representation of these distances, and that fails with 3 samples.

I think I would need to dig into the code a bit to fix it, I'm afraid.

@alanaleah89 your problem is at least easier to fix, if you know how to dig into R functions. You could try something like:

debug(calc_pairwise_mmds)
qc_obj = calc_pairwise_mmds(qc_dt, qc_names, annots_disc = NULL, annots_cont = annots_cont, n_cores = 5, one_group_only = TRUE)

then step through the function one line at a time. At some point, you'll get to the line n_nhbrs = 5. After you've done that line, run n_nhbrs = 3, then step through the rest. The issue here is that SampleQC is trying to run UMAP on your 6 samples, using the 5 nearest neighbours, and that doesn't really make any sense...

Hope this helps!
Will

[wmacnair]

Also, I think there's a slight misunderstanding of the annots_disc parameter. The idea with this is that when you have a big dataset with lots of different samples, it might be helpful to see if there is any relationship between QC metrics and sample variables, e.g. batch.

To illustrate this:

• annots_disc = c('SampleName') doesn't really add anything, because the model already stores the sample names.
• annots_disc = c("cell_id", "sample_id", "condition") doesn't make sense, because the cell_id labels are at the cell rather than sample level.
• annots_disc = "condition" and annots_disc = "batch" both make perfect sense 🥳
• annots_disc = NULL is also fine

Best
Will