Repair and optimize PsiNorm()

[hpages]

Hi,

A recent change in DelayedArray (in version 0.33.5) broke PsiNorm() on a sparse DelayedMatrix object and on a SE/SCE object with counts represented as a sparse DelayedMatrix object. See https://bioconductor.org/checkResults/3.21/bioc-LATEST/scone/. Sorry for that!

This commit fixes the problem and also introduces two small improvements to PsiNorm():

• makes PsiNorm() work on a SparseMatrix object;
• makes PsiNorm() slightly faster by reducing the number of matrix transpositions from 4 to 2.

For example:

library(scone)

library(DelayedArray)
x <- DelayedArray(poissonSparseMatrix(10, 8))
# Broken in scone 1.31.0 (due to recent change in DelayedArray) but works again in scone 1.31.1
PsiNorm(x)

library(TENxPBMCData)
sce <- TENxPBMCData("pbmc4k")
# Also broken in scone 1.31.0 (due to recent change in DelayedArray) but works again in scone 1.31.1
PsiNorm(sce)

Other improvements:

x0 <- poissonSparseMatrix(5000, 77000)
a0 <- as.matrix(x0)

PsiNorm(a0)  # 50% speedup compared to scone 1.30.0

PsiNorm(x0)  # NEW (not supported in scone 1.30.0), 10x faster than 'PsiNorm(a0)'

sce <- SingleCellExperiment(list(counts=x0))
PsiNorm(sce)  # NEW (not supported in scone 1.30.0)

Best,
H.

[drisso]

Thank you @hpages