Return variable features directly if you have set them yourself

[Gesmira]

This has a corresponding fix in seurat-private: PR. s/o @longmanz for identifying and helping fix this issue.

The issue occurred when we tried to pull out VariableFeatures from an object in which we first ran FindVariableFeatures(), and then set our own VariableFeatures after. If we tried to pull out the re-set variable features, it pulled out the features previously calculated with FindVariableFeatures before

pbmc3k_v2 = FindVariableFeatures(pbmc3k)
VariableFeatures(pbmc3k_v2) <- rownames(pbmc3k_v2)[1:100]
head(pbmc3k_v2[["RNA"]]@meta.data)
# won't return the ones you've set 
head(VariableFeatures(pbmc3k_v2[["RNA"]], nfeatures = 2000))

The issue was that we have a check that we skip to the calculated variable features if nfeatures != Inf or the # of variable features. The solution is to almost always return the variable features directly if you have set them yourself.

Documented more here: https://github.com/orgs/satijalab/projects/1/views/1?pane=issue&itemId=51523433

[longmanz]

I have tested the fix and it works well and as expected from my end. However, do we want to allow users to use the nfeatures = n parameter in VariableFeatures()? Now this does not work and VariableFeatures() will simply output all genes if users set VariableFeatures manually.

A naive modification would be:

if (isTRUE(x = simplify) & (is.null(x = layer) || any(is.na(x = layer))) &
         (is.infinite(x = nfeatures) || length(x = var.features) <=
          nfeatures)) {
       return(var.features) 
    } else if (isTRUE(x = simplify) & (is.null(x = layer) || any(is.na(x = layer))) &
         (is.infinite(x = nfeatures) || length(x = var.features) >
          nfeatures)) {
       return(var.features[1:nfeatures])
}