Assemble MAE/MASE reads with new2(check = FALSE); guard with validObject tests

readParquet() now builds multi_assay_experiment and
multi_assay_spatial_experiment objects directly via S4Vectors::new2(check =
FALSE) instead of the MultiAssayExperiment() / MultiAssaySpatialExperiment()
constructors. Those constructors re-derive sample-name set equality at
construction time with a locale-collation sort over the full, cell-grain
sampleMap, which dominates read time on large products (tens of millions of
sampleMap rows -- a cell-resolved spatial atlas took ~22 min to read, over 60%
of it in that redundant sort). The datapackage already carries writeParquet()'s
write-time integrity guarantee (a DuckDB anti-join proving every sampleMap
'primary' is a subjects rowname), so the constructor-time re-check is redundant;
the assembled object still passes validObject(). The direct assembly also skips
harmonization, so unreferenced colData rows are retained rather than dropped on
read (write with subset_subjects_to_referenced upstream to avoid them).

Because the read path no longer runs constructor-time validity, the MAE/MASE
round-trip tests now assert validObject() on every readParquet() result -- in
the two MAE round-trips, the MASE datapackage-schema round-trip, and the two
spatial fixture helpers (makeLazySpatialMASE, .lazyPointsMASE) covering every
spatial-query test. This closes the gap that would otherwise let a
structurally-invalid readParquet()-assembled object through.

Author: aboyounp-gne

DESCRIPTION

@@ -1,5 +1,5 @@
 Package: BiocDuckDB
-Version: 0.99.10
+Version: 0.99.11
 Date: 2026-07-28
 Title: Bioconductor DuckDB Integration and High-Level I/O
 Description:

NAMESPACE

@@ -95,6 +95,7 @@ importClassesFrom(MultiAssayExperiment,ExperimentList)
 importClassesFrom(MultiAssayExperiment,MultiAssayExperiment)
 importClassesFrom(MultiAssaySpatialExperiment,MultiAssaySpatialExperiment)
 importClassesFrom(MultiAssaySpatialExperiment,PointsLayerList)
+importClassesFrom(MultiAssaySpatialExperiment,RasterLayerList)
 importClassesFrom(MultiAssaySpatialExperiment,ShapesLayerList)
 importClassesFrom(S4Vectors,DFrame)
 importClassesFrom(S4Vectors,DataFrame)

NEWS.md

@@ -1,3 +1,30 @@
+# BiocDuckDB 0.99.11
+
+## Enhancements
+
+- `readParquet()` now assembles `multi_assay_experiment` and
+  `multi_assay_spatial_experiment` objects directly with
+  `S4Vectors::new2(check = FALSE)` instead of calling the
+  `MultiAssayExperiment()` / `MultiAssaySpatialExperiment()` constructors. Those
+  constructors re-derive sample-name set equality at construction time via a
+  locale-collation sort over the full, cell-grain `sampleMap`, which dominates
+  read time on large products (tens of millions of `sampleMap` rows -- e.g. a
+  cell-resolved spatial atlas took ~22 min to read, over 60% of it in that
+  redundant sort). The datapackage already carries `writeParquet()`'s write-time
+  integrity guarantee (a DuckDB anti-join proving every `sampleMap` `primary` is
+  a subjects rowname), so the constructor-time re-check is redundant; the
+  assembled object still passes `validObject()`. The direct assembly also skips
+  harmonization, so unreferenced `colData` rows are retained rather than dropped
+  on read (write with `subset_subjects_to_referenced` upstream to avoid them).
+
+## Testing
+
+- The `multi_assay_experiment` and `multi_assay_spatial_experiment` round-trip
+  tests now assert `validObject()` on every object returned by `readParquet()`.
+  Because the read path assembles with `new2(check = FALSE)` (no constructor-time
+  validity), these explicit checks guard against a `readParquet()`-assembled
+  object being structurally invalid.
+
 # BiocDuckDB 0.99.10
 
 ## Enhancements

R/readParquet.R

@@ -699,6 +699,41 @@ function(path,
 ### MultiAssayExperiment objects
 ###
 
+#' @importClassesFrom MultiAssayExperiment MultiAssayExperiment
+#' @importClassesFrom S4Vectors DataFrame
+#' @importFrom MultiAssayExperiment ExperimentList
+#' @importFrom S4Vectors new2
+.newMAE <- function(experiments, colData, sampleMap, metadata = list()) {
+    if (is.null(metadata)) {
+        metadata <- list()
+    }
+    new2("MultiAssayExperiment",
+         ExperimentList = ExperimentList(experiments),
+         colData        = colData,
+         sampleMap      = as(sampleMap, "DataFrame"),
+         metadata       = metadata,
+         check          = FALSE)
+}
+
+#' @importClassesFrom MultiAssaySpatialExperiment MultiAssaySpatialExperiment
+#' @importClassesFrom MultiAssaySpatialExperiment RasterLayerList
+#' @importClassesFrom MultiAssaySpatialExperiment PointsLayerList
+#' @importClassesFrom MultiAssaySpatialExperiment ShapesLayerList
+#' @importFrom S4Vectors new2
+.newMASE <-
+function(mae, images, labels, points, shapes, imgData, spatialMap)
+{
+    new2("MultiAssaySpatialExperiment",
+         mae,
+         images     = as(images, "RasterLayerList"),
+         labels     = as(labels, "RasterLayerList"),
+         points     = as(points, "PointsLayerList"),
+         shapes     = as(shapes, "ShapesLayerList"),
+         imgData    = imgData,
+         spatialMap = spatialMap,
+         check      = FALSE)
+}
+
 #' @importFrom MultiAssayExperiment MultiAssayExperiment
 #' @importFrom stats setNames
 .readParquetMAE <- function(path, package, ...) {
@@ -735,10 +770,8 @@ function(path,
     metadata <- .deserializeMetadata(package[["annotations"]], resources, path, ...)
 
     # MultiAssayExperiment
-    MultiAssayExperiment(experiments,
-                         colData = subjects,
-                         sampleMap = sample_map,
-                         metadata = metadata)
+    .newMAE(experiments, colData = subjects, sampleMap = sample_map,
+            metadata = metadata)
 }
 
 ### - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
@@ -896,16 +929,7 @@ function(path,
         spatial_map <- DataFrame(spatial_map, check.names = FALSE)
     }
 
-    MultiAssaySpatialExperiment::MultiAssaySpatialExperiment(
-        experiments = experiments(mae),
-        colData = colData(mae),
-        sampleMap = sampleMap(mae),
-        images = images,
-        labels = labels,
-        points = points,
-        shapes = shapes,
-        imgData = img_data,
-        spatialMap = spatial_map,
-        metadata = metadata(mae)
-    )
+    # MultiAssaySpatialExperiment
+    .newMASE(mae, images = images, labels = labels, points = points,
+             shapes = shapes, imgData = img_data, spatialMap = spatial_map)
 }

tests/testthat/setup.R

@@ -299,5 +299,9 @@ makeLazySpatialMASE <- function(path = NULL) {
     if (is.null(path))
         path <- file.path(tempdir(), paste0("mase_lazy_", sample.int(1e6, 1L)))
     BiocDuckDB::writeParquet(mase, path)
-    BiocDuckDB::readParquet(path)
+    out <- BiocDuckDB::readParquet(path)
+    # readParquet assembles the MASE with new2(check = FALSE); assert the
+    # write-trusted fast path still yields a valid object (guards every caller).
+    expect_true(validObject(out))
+    out
 }

tests/testthat/test-datapackage-schema.R

@@ -222,6 +222,7 @@ test_that("MASE conforms spatialMap via an spatial_element_registry + monomorphi
 
     # Round-trip: the internal spine does not leak back into the object model.
     m2 <- readParquet(tmpdir)
+    expect_true(validObject(m2))
     sm2 <- MultiAssaySpatialExperiment::spatialMap(m2)
     expect_false(any(c("__element__", "__index__") %in% colnames(sm2)))
     expect_true(all(c("assay", "colname", "element_type", "region",

tests/testthat/test-parquet-roundtrip.R

@@ -351,6 +351,7 @@ test_that("MultiAssayExperiment with nested experiments works", {
 
     # Check structure
     expect_s4_class(mae2, "MultiAssayExperiment")
+    expect_true(validObject(mae2))
     expect_identical(names(experiments(mae2)), c("RNA", "Protein"))
     expect_identical(length(experiments(mae2)), 2L)
 
@@ -393,6 +394,7 @@ test_that("MultiAssayExperiment with flat array-like objects works", {
 
     # Check structure
     expect_s4_class(mae2, "MultiAssayExperiment")
+    expect_true(validObject(mae2))
     expect_identical(names(experiments(mae2)), c("Dataset1", "Dataset2"))
     expect_identical(length(experiments(mae2)), 2L)
 

tests/testthat/test-spatial-query.R

@@ -32,7 +32,11 @@
                 scaled = list(type = "scale", scale = c(2, 2))))))
     path <- file.path(tempdir(), paste0("mase_q_", sample.int(1e6, 1L)))
     BiocDuckDB::writeParquet(mase, path)
-    BiocDuckDB::readParquet(path)
+    out <- BiocDuckDB::readParquet(path)
+    # readParquet assembles the MASE with new2(check = FALSE); assert the
+    # write-trusted fast path still yields a valid object (guards every caller).
+    expect_true(validObject(out))
+    out
 }
 
 test_that("per-element transforms round-trip and expose coordinate systems", {