Switch to beachmat to check each row for missing values.

This allows us to get rid of a lot of the BPPARAM= machinery.
SingleR-inc · Dec 15, 2024 · afd3d99 · afd3d99
1 parent 80b9923
commit afd3d99
Show file tree

Hide file tree

Showing 9 changed files with 20 additions and 40 deletions.
diff --git a/DESCRIPTION b/DESCRIPTION
@@ -1,7 +1,7 @@
 Package: SingleR
 Title: Reference-Based Single-Cell RNA-Seq Annotation
-Version: 2.9.2
-Date: 2024-11-24
+Version: 2.9.3
+Date: 2024-12-14
 Authors@R: c(person("Dvir", "Aran", email="[email protected]", role=c("aut", "cph")),
     person("Aaron", "Lun", email="[email protected]", role=c("ctb", "cre")),
     person("Daniel", "Bunis", role="ctb"),
@@ -23,7 +23,7 @@ Imports:
     stats,
     utils,
     Rcpp,
-    beachmat (>= 2.21.1)
+    beachmat (>= 2.23.5)
 LinkingTo:
     Rcpp,
     beachmat,

diff --git a/NAMESPACE b/NAMESPACE
@@ -33,10 +33,7 @@ importFrom(BiocNeighbors,AnnoyParam)
 importFrom(BiocNeighbors,KmknnParam)
 importFrom(BiocNeighbors,defineBuilder)
 importFrom(BiocParallel,SerialParam)
-importFrom(BiocParallel,bpisup)
 importFrom(BiocParallel,bpnworkers)
-importFrom(BiocParallel,bpstart)
-importFrom(BiocParallel,bpstop)
 importFrom(DelayedArray,DelayedArray)
 importFrom(DelayedArray,colsum)
 importFrom(DelayedArray,getAutoBPPARAM)
@@ -59,6 +56,7 @@ importFrom(S4Vectors,selfmatch)
 importFrom(SummarizedExperiment,SummarizedExperiment)
 importFrom(SummarizedExperiment,assay)
 importFrom(beachmat,initializeCpp)
+importFrom(beachmat,tatami.row.nan.counts)
 importFrom(methods,as)
 importFrom(methods,is)
 importFrom(stats,mad)

diff --git a/R/SingleR.R b/R/SingleR.R
@@ -102,14 +102,9 @@ SingleR <- function(
     BNPARAM = NULL,
     BPPARAM=SerialParam()) 
 {
-    if (!bpisup(BPPARAM) && !is(BPPARAM, "MulticoreParam")) {
-        bpstart(BPPARAM)
-        on.exit(bpstop(BPPARAM))
-    }
-
     # We have to do all this row-subsetting at the start before trainSingleR,
     # otherwise 'test.genes' won't match up to the filtered 'test'.
-    test <- .to_clean_matrix(test, assay.type.test, check.missing.test, msg="test", BPPARAM=BPPARAM)
+    test <- .to_clean_matrix(test, assay.type.test, check.missing.test, msg="test", num.threads=num.threads)
 
     tmp.ref <- ref
     if (!is.list(tmp.ref) || is.data.frame(tmp.ref)) {

diff --git a/R/classifySingleR.R b/R/classifySingleR.R
@@ -104,7 +104,7 @@ classifySingleR <- function(
     num.threads = bpnworkers(BPPARAM),
     BPPARAM=SerialParam()) 
 {
-    test <- .to_clean_matrix(test, assay.type, check.missing=FALSE, msg="test", BPPARAM=BPPARAM)
+    test <- .to_clean_matrix(test, assay.type, check.missing=FALSE, msg="test", num.threads=num.threads)
 
     solo <- .is_solo(trained)
     if (solo) { 

diff --git a/R/combineRecomputedResults.R b/R/combineRecomputedResults.R
@@ -120,7 +120,7 @@ combineRecomputedResults <- function(
     num.threads = bpnworkers(BPPARAM),
     BPPARAM=SerialParam())
 {
-    test <- .to_clean_matrix(test, assay.type=assay.type.test, check.missing=FALSE, msg="test", BPPARAM=BPPARAM)
+    test <- .to_clean_matrix(test, assay.type=assay.type.test, check.missing=FALSE, msg="test", num.threads=num.threads)
 
     # Applying the sanity checks.
     stopifnot(length(results) == length(trained))

diff --git a/R/getClassicMarkers.R b/R/getClassicMarkers.R
@@ -54,18 +54,13 @@
 #' @importFrom utils relist
 #' @importFrom beachmat initializeCpp
 getClassicMarkers <- function(ref, labels, assay.type="logcounts", check.missing=TRUE, de.n=NULL, num.threads=bpnworkers(BPPARAM), BPPARAM=SerialParam()) { 
-    if (!bpisup(BPPARAM) && !is(BPPARAM, "MulticoreParam")) {
-        bpstart(BPPARAM)
-        on.exit(bpstop(BPPARAM))
-    }
-
     if (!.is_list(ref)) { 
         ref <- list(ref)
         labels <- list(labels)
     }
 
     for (i in seq_along(ref)) {
-        ref[[i]] <- .to_clean_matrix(ref[[i]], assay.type, check.missing, msg="ref", BPPARAM=BPPARAM)
+        ref[[i]] <- .to_clean_matrix(ref[[i]], assay.type, check.missing, msg="ref", num.threads=num.threads)
     }
 
     # Setting up references.

diff --git a/R/plotMarkerHeatmap.R b/R/plotMarkerHeatmap.R
@@ -81,7 +81,7 @@ plotMarkerHeatmap <- function(
     BPPARAM = SerialParam(),
     ...) 
 {
-    test <- .to_clean_matrix(test, assay.type, check.missing=FALSE)
+    test <- .to_clean_matrix(test, assay.type, check.missing=FALSE, num.threads=num.threads)
     config <- configureMarkerHeatmap(
         results, 
         test,
@@ -129,7 +129,7 @@ configureMarkerHeatmap <- function(
     order.by.summary="min.rank",
     num.threads=1)
 {
-    test <- .to_clean_matrix(test, assay.type, check.missing=FALSE)
+    test <- .to_clean_matrix(test, assay.type, check.missing=FALSE, num.threads=num.threads)
     all.markers <- metadata(results)$de.genes[[label]]
 
     if (use.pruned) {

diff --git a/R/trainSingleR.R b/R/trainSingleR.R
@@ -195,7 +195,7 @@
 #' @export
 #' @importFrom S4Vectors List isSingleString metadata metadata<-
 #' @importFrom BiocNeighbors defineBuilder AnnoyParam KmknnParam
-#' @importFrom BiocParallel SerialParam bpisup bpstart bpstop
+#' @importFrom BiocParallel SerialParam
 #' @importFrom beachmat initializeCpp
 #' @importFrom S4Vectors List
 #' @importFrom SummarizedExperiment assay
@@ -243,15 +243,10 @@ trainSingleR <- function(
         }
     }
 
-    if (!bpisup(BPPARAM) && !is(BPPARAM, "MulticoreParam")) {
-        bpstart(BPPARAM)
-        on.exit(bpstop(BPPARAM))
-    }
-
     output <- vector("list", length(ref))
     names(output) <- names(ref)
     for (l in seq_along(ref)) {
-        curref <- .to_clean_matrix(ref[[l]], assay.type, check.missing, msg="ref", BPPARAM=BPPARAM)
+        curref <- .to_clean_matrix(ref[[l]], assay.type, check.missing, msg="ref", num.threads=num.threads)
 
         curlabels <- as.character(labels[[l]])
         stopifnot(length(curlabels) == ncol(curref))
@@ -274,7 +269,7 @@ trainSingleR <- function(
         )
 
         if (aggr.ref) {
-            aggr <- do.call(aggregateReference, c(list(ref=quote(curref), label=curlabels, check.missing=FALSE, BPPARAM=BPPARAM), aggr.args))
+            aggr <- do.call(aggregateReference, c(list(ref=quote(curref), label=curlabels, check.missing=FALSE, num.theads=num.threads), aggr.args))
             curref <- assay(aggr)
             curlabels <- aggr$label
         }

diff --git a/R/utils.R b/R/utils.R
@@ -1,8 +1,9 @@
+#' @importFrom beachmat initializeCpp tatami.row.nan.counts
 #' @importFrom SummarizedExperiment assay
 #' @importFrom DelayedMatrixStats rowAnyNAs
 #' @importClassesFrom SummarizedExperiment SummarizedExperiment
-#' @importFrom DelayedArray DelayedArray setAutoBPPARAM getAutoBPPARAM
-.to_clean_matrix <- function(x, assay.type, check.missing, msg="x", BPPARAM=SerialParam()) {
+#' @importFrom DelayedArray DelayedArray
+.to_clean_matrix <- function(x, assay.type, check.missing, msg="x", num.threads=1) {
     if (is.null(rownames(x)) && nrow(x)) { # zero-length matrices have NULL dimnames.
         stop(sprintf("'%s' must have row names", msg))
     }
@@ -20,15 +21,11 @@
 
     # Stripping out genes with NA's from 'x'.
     if (check.missing) {
-        old <- getAutoBPPARAM()
-        setAutoBPPARAM(BPPARAM)
-        on.exit(setAutoBPPARAM(old))
-
-        y <- DelayedArray(x)
-        discard <- rowAnyNAs(y)
-        if (any(discard)) {
+        ptr <- initializeCpp(x)
+        keep <- tatami.row.nan.counts(ptr, num.threads=num.threads) == 0
+        if (!all(keep)) {
             # Returning a DelayedArray to avoid making an actual subset.
-            x <- y[!discard,,drop=FALSE]
+            x <- DelayedArray(y)[keep,,drop=FALSE]
         }
     }