Changes

Additions

• Added scSLIDE (Single-cell Sample-Level Inference using Density Estimation) to Seurat, enabling sample-level analysis of single-cell data
  • Added PrepareSampleObject to select landmark cells and map every cell to its nearest landmarks
  • Added SketchDataByGroup to sketch cells within metadata groups using uniform or leverage-score sampling, with configurable cell counts per group and layer
  • Added QuickCorTest to select genes correlated with a response variable, optionally within metadata groups
  • Added RunPLS to compute a partial least squares (PLS) dimensional reduction, whose components capture the variation in expression most predictive of one or more sample- or cell-level responses (supports plsr, cppls and spls)
  • Added FindmmNN and FindsmNN to map cells to landmarks using weighted combinations of reductions or a single reduction, respectively
  • Added GenerateSampleObject to summarize landmark assignments per sample into a Seurat object for downstream sample-level analysis
  • Added NormalizeChiSquared to normalize raw landmark-by-sample counts as deviations from expected counts
  • Added RunDiffusionMap to compute a diffusion map reduction with destiny, including support for a precomputed distance matrix and eigenvalue-weighted embeddings
  • Added TrajDETest to test expression changes along a continuous trajectory using negative binomial regression with glmGamPoi
  • Added PlotAnnotatedHeatmap to display expression heatmaps with cells or samples ordered by metadata variables and optional categorical or continuous annotation bars
• Added ReadAtera and LoadAtera to read 10x Genomics Atera spatial data, including expression and control counts, Azimuth cell type annotations, cell centroids, cell/nucleus segmentations, transcript coordinates, and morphology images
  • Added support for gene selection when reading molecules, feature-type selection when reading counts, BPCells-backed counts caching, and parallel Atera reads using the active future::plan()
  • Added LoadAteraMolecules, LoadAteraSegmentations, and LoadAteraCodewords to load additional transcript coordinates, segmentation polygons, and per-codeword coordinates into existing Atera objects
• Added gene and region selection to ReadXenium and LoadXenium for molecule coordinates

Fixes

• Fixed LeverageScore (and so SketchData) on objects with fewer than 50 features or cells by using the number of components already computed from the object's size
• Updated LeverageScore (and so SketchData) to scale better on BPCells on-disk data, by computing the random projection in blocks of cells instead of transposing the whole matrix. Results for a given seed differ from previous versions
• Fixed LeverageScore on BPCells on-disk data deleting unrelated files from the session temporary directory and leaving behind a full on-disk copy of the input matrix
• Fixed LeverageScore densifying the whole layer when retrying after a failure, and reporting that no zero-variance features were found even when it had removed some
• Fixed ProjectData and TransferSketchLabels not preserving factor levels and ordering in projected labels
• Fixed SCT query projection using the resolved reference variance floor instead of the original "umi_median" setting, which could reduce anchors in FindTransferAnchors and cause TransferData to fail
• Improved table reading in Read10X, ReadMtx, LoadSTARmap, LoadCurioSeeker, and ReadSlideSeq, and coordinate handling in SpatialPlot
• Fixed Load10X_Spatial with custom assay and slice names and uppercase feature names, and cell filtering in Read10X_Coordinates, Read10X_Segmentations, and Read10X_HD_GeoJson
• Fixed ImageDimPlot and ImageFeaturePlot with FOV and boundary names containing underscores
• Fixed default orientation in ImageDimPlot and ImageFeaturePlot for spatial FOVs, including older Xenium/Atera objects: x is horizontal and y increases downward; stored coordinates are unchanged
• Fixed transcript quality filtering and count reader error reporting in ReadXenium
• Fixed LoadXenium molecule labels to match assay feature names when underscores or pipes are replaced with dashes
• Reduced data exported to parallel workers by passing selected inputs directly
  • Per-object inputs in PrepSCTIntegration, FindIntegrationAnchors, MapQueryData, and BridgeCellsRepresentation
  • Feature chunks in RunMarkVario
  • Cell names per field of view for segmentations and cell metadata rows for bounding boxes in ReadVizgen

Deprecations

• Soft deprecation of flip.xy in LoadXenium
  • Existing calls with flip.xy = TRUE will retain their behavior (requiring SeuratObject >= v5.3.0 to do so)