.testMarkerSC() failing when mean1 and/or mean2 is NA during call to getMarkerFeatures()

[wuv21]

Hi!

Been running into an issue similar to #562 and #483 with the same error output, but I have been running ArchR v1.0.2 (commit 48dccf) which was the solution to both those questions. As such, I believed that the error was related to something else.

By debugging through the getMarkerFeatures() call (see below for my function call):

tmp <- getMarkerFeatures(
  ArchRProj = proj,
  useMatrix = "MotifMatrix",
  groupBy = "condensedCluster",
  testMethod = "wilcoxon",
  bias = c("TSSEnrichment", "log10(nFrags)"),
  useGroups = "CD4_TRUE",
  bgdGroups = "CD4_FALSE",
  maxCells = 1000,
  verbose = TRUE,
  useSeqnames = "z"
)

I noticed that one of the rows in pairwiseDF (see source below) contained NA values for both mean1 and mean2 (small section of data frame shown below). The row "f156" is the only row with this issue. All other rows were ok in the rest of pairwiseDF.

ArchR/R/MarkerFeatures.R

Lines 422 to 426 in 48dccf3

	idxFilter <- rowSums(pairwiseDF[,c("mean1","mean2")]) != 0
	pairwiseDF$fdr <- NA
	pairwiseDF$fdr[idxFilter] <- p.adjust(pairwiseDF$pval[idxFilter], method = "fdr")
	pairwiseDF <- pairwiseDF[rownames(featureDF), , drop = FALSE]
	pairwiseDF

     log2Mean     log2FC          fdr         pval     mean1      mean2  n       auc
f155 3.089913        NaN 7.740502e-02 5.694163e-03 0.6838235 -0.2663543 36 0.6898148
f156 0.000000 0.00000000 2.650317e-10 3.046341e-13        NA         NA 36 0.0000000
f157 3.705964 0.08330494 9.909369e-01 9.147955e-01 0.3451817  0.3242560 36 0.4922840

Since mean1 and mean2 were NA, idxfilter contained a vector of both boolean values and a NA that was associated with row "f156". I think the fix below would solve this issue:

  idxFilter <- rowSums(pairwiseDF[, c("mean1", "mean2")]) != 0 & rowSums(is.na(pairwiseDF[, c("mean1", "mean2")])) == 0

Thanks!

Edit: updated my fix; didn't realize Github changed a piece of code into a link.
Edit2: updated fix again.