Merge pull request #8271 from bnprks/develop

Improve BPCells compatibility (JackStraw + RunPCA)

DESCRIPTION

@@ -1,5 +1,5 @@
 Package: Seurat
-Version: 5.1.0.9011
+Version: 5.1.0.9012
 Title: Tools for Single Cell Genomics
 Description: A toolkit for quality control, analysis, and exploration of single cell RNA sequencing data. 'Seurat' aims to enable users to identify and interpret sources of heterogeneity from single cell transcriptomic measurements, and to integrate diverse types of single cell data. See Satija R, Farrell J, Gennert D, et al (2015) <doi:10.1038/nbt.3192>, Macosko E, Basu A, Satija R, et al (2015) <doi:10.1016/j.cell.2015.05.002>, Stuart T, Butler A, et al (2019) <doi:10.1016/j.cell.2019.05.031>, and Hao, Hao, et al (2020) <doi:10.1101/2020.10.12.335331> for more details.
 Authors@R: c(

NEWS.md

@@ -14,6 +14,8 @@
 - Fixed `RunPCA` to avoid converting `BPCells` matrices into dense matrices - significantly reduces the function's memory usage when running on `BPCells` matrices
 - Added `features` parameter to `LeverageScore` and `SketchData`
 - Updated `SketchData`'s `ncells` parameter to accept integer vector
+- Updated `JackStraw` to support `BPCells` matrices
+- Updated `RunPCA` to use the `BPCells`-provided SVD solver on `BPCells` matrices
 
 # Seurat 5.1.0 (2024-05-08)
 

R/dimensional_reduction.R

@@ -865,19 +865,22 @@ RunPCA.default <- function(
   }
  if (inherits(x = object, what = 'matrix')) {
    RowVar.function <- RowVar
+   svd.function <- irlba
  } else if (inherits(x = object, what = 'dgCMatrix')) {
    RowVar.function <- RowVarSparse
+   svd.function <- irlba
  } else if (inherits(x = object, what = 'IterableMatrix')) {
    RowVar.function <- function(x) {
      return(BPCells::matrix_stats(
        matrix = x,
        row_stats = 'variance'
      )$row_stats['variance',])
-     }
+    }
+    svd.function <- function(A, nv, ...) BPCells::svds(A=A, k = nv)
  }
   if (rev.pca) {
     npcs <- min(npcs, ncol(x = object) - 1)
-    pca.results <- irlba(A = object, nv = npcs, ...)
+    pca.results <- svd.function(A = object, nv = npcs, ...)
     total.variance <- sum(RowVar.function(x = t(x = object)))
     sdev <- pca.results$d/sqrt(max(1, nrow(x = object) - 1))
     if (weight.by.var) {
@@ -891,7 +894,7 @@ RunPCA.default <- function(
     total.variance <- sum(RowVar.function(x = object))
     if (approx) {
       npcs <- min(npcs, nrow(x = object) - 1)
-      pca.results <- irlba(A = t(x = object), nv = npcs, ...)
+      pca.results <- svd.function(A = t(x = object), nv = npcs, ...)
       feature.loadings <- pca.results$v
       sdev <- pca.results$d/sqrt(max(1, ncol(object) - 1))
       if (weight.by.var) {
@@ -2335,7 +2338,7 @@ JackRandom <- function(
     rand.genes <- sample(x = rownames(x = scaled.data), size = 3)
   }
   data.mod <- scaled.data
-  data.mod[rand.genes, ] <- MatrixRowShuffle(x = scaled.data[rand.genes, ])
+  data.mod[rand.genes, ] <- MatrixRowShuffle(x = as.matrix(scaled.data[rand.genes, ]))
   temp.object <- RunPCA(
     object = data.mod,
     assay = "temp",