Error in pgx.computePGX when working on small datasets

[phisanti]

Error in pgx.computePGX when working on small datasets

[ivokwee]

I would debug the error. Sometimes people have just 500-1000 genes. So it is a good edge case.

[phisanti]

The error seems to come from the function compute_testGenesets in compute2-geneset.R. Apparently, there is a hard limit to gene set size. If a gene set has lower than 15 genes, then it won't select any column for further analysis. Here is the code:

  ## filter gene sets on size
  cat("Filtering gene sets on size...\n")
  gmt.size <- Matrix::colSums(G != 0)
  size.ok <- (gmt.size >= 15 & gmt.size <= 400)
  G <- G[, which(size.ok)]

Here we have two options:

1. We can reduce the limit or making it relative to the input data.
2. Try some kind of try-catch method so that the script can continue but the geneset will be skipped.

Any suggestion @ivokwee @ncullen93?

[ncullen93]

Yeah a try-catch method would be best. I'm not sure if this error occurs in the omicsplayground platform or if this issue is caught before this code gets run. This is a good example of why all data catching code should be in playbase and not in omicsplayground. Looping in @ivokwee on this.

[phisanti]

Ok, taking on account that last comment, I came up with the following solution:

  # If dataset is too small that size.ok == 0, then select top 100
  if (sum(size.ok) == 0) {
    top_100gs <- head(sort(gmt.size,decreasing = TRUE), 100)
    size.ok <- names(gmt.size) %in% names(top_100gs)
  }
  

Injecting this code in compute_testGenesets after calculating size.ok should solve the issue. This way, there will always be at least 100 genesets. Let me know what you think, @ncullen93 @ivokwee.

[ivokwee]

Yes that's a good solution. But please test.

[phisanti]

I can confirm the edit works in the Playbase tests as well as in the omicsplayground