Skip to contents

Changelog

Source: NEWS.md

srtdisk 0.3.1

Release Date: 2026-02-25

Bug Fixes

  • Fixed CSR sparse matrix loading failure in ReadSparseMatrix() that caused 'dims' must contain all (i,j) pairs errors when loading large h5ad-converted h5Seurat files. The dims argument to Matrix::sparseMatrix() was incorrectly swapped as (ncols, nrows) instead of (nrows, ncols), and the unnecessary transpose has been removed.
  • Added direct h5ad-to-RDS conversion path in Convert() so that Convert("file.h5ad", dest = "file.rds") works without a manual intermediate h5Seurat step.

Validated on mouse atlas (356K cells, 16.8K genes) and human atlas (1.04M cells, 17.1K genes).

srtdisk 0.3.0

Release Date: 2026-02-09

Highlights

MuData (h5mu) Multimodal Support

srtdisk now supports the MuData/h5mu format for multimodal single-cell data:

  • LoadH5MU(): Load h5mu files as multimodal Seurat objects with automatic modality-to-assay name mapping (rna->RNA, prot->ADT, atac->ATAC, etc.)
  • SaveH5MU(): Export multimodal Seurat objects to h5mu format with spatial data preservation and conflict-free modality naming
  • as.h5mu(): Convenience alias for SaveH5MU
  • H5MU conversion pathways: Convert() now supports h5mu as both source and destination format (h5mu<->h5Seurat, h5mu->h5ad)
  • Requires MuDataSeurat package (optional dependency)

Direct h5ad Loading with LoadH5AD()

New LoadH5AD() function provides direct h5ad-to-Seurat conversion without the intermediate h5Seurat step:

  • Reads sparse (CSR) and dense expression matrices
  • Preserves categorical metadata as R factors
  • Restores dimensional reductions (PCA, UMAP, tSNE)
  • Handles feature metadata, neighbor graphs, and spatial data
  • Supports raw counts from raw/X and additional layers

New Vignettes

  • Direct H5AD Loading (direct-h5ad-loading): Showcases LoadH5AD() with real data from CellxGene and SeuratData
  • Multimodal H5MU (multimodal-h5mu): Demonstrates LoadH5MU()/SaveH5MU() round-trip with CITE-seq data

New Internal Utilities

Code Simplification

  • Removed unused UtilsSparseMatrix.R (ReadSparseMatrixDims, ValidateSparseMatrixDims, CreateSparseMatrixSafe)
  • Removed unused GetSeuratSlotMapping() and ValidateSlotMapping() from V5Compatibility.R
  • Removed fragile PatchH5DMethods() monkey-patching of hdf5r’s H5D [ method
  • Fixed GetScaleFactors() to extract actual Visium scale factors via scalefactors() instead of returning hardcoded values

Preserved from v0.2.2

All v0.2.2 improvements are preserved in this release: - Unified Convert() interface accepting Seurat objects directly - Exported SeuratToH5AD() convenience function - Column name sanitization (SanitizeColumnName/SanitizeColumnNames) - Correct categorical encoding (no +1L regression) - Nullable dtype handling (FlattenNullable) - Variable features writing in V5 compatibility layer - standardize parameter for scanpy-compatible naming

srtdisk 0.2.2

Release Date: 2026-02-06

Highlights

Unified Convert() Interface for Seurat Objects

Convert() now accepts Seurat objects directly as the source argument, enabling one-step conversion to any supported format:

Convert(seurat_obj, dest = "output.h5ad")     # Seurat -> h5ad
Convert(seurat_obj, dest = "output.loom")     # Seurat -> Loom
Convert(seurat_obj, dest = "output.h5Seurat") # Seurat -> h5Seurat

This eliminates the need for the two-step SaveH5Seurat() + Convert() workflow when starting from a Seurat object.

Exported SeuratToH5AD()

SeuratToH5AD() is now a fully exported convenience function (previously @keywords internal). It delegates to Convert.Seurat internally and remains the simplest way to convert a Seurat object to h5ad format.

Bug Fixes

  • Fixed var/X dimension mismatch in h5Seurat to h5ad conversion: var now uses the actual X matrix dimensions instead of metadata feature counts, which could differ when features were filtered from data layers but not from metadata
  • Fixed raw/var dimension mismatch with raw/X using the same approach
  • Fixed roxygen2::roxygenize() failure caused by a plain IsFactor() function in ReadH5.R overriding the S4 generic from TestObject.R

srtdisk 0.2.1

Release Date: 2026-02-05

Highlights

Correct Gene Set in h5Seurat to h5ad Conversion

Fixed a critical issue where scale.data (containing only ~2,000 variable features) was prioritized as X during h5Seurat to h5ad conversion, causing var to contain only variable genes instead of the full gene set. The conversion now correctly uses data (normalized, all genes) as X and counts as raw/X, matching standard AnnData conventions. scale.data is only used as a last-resort fallback.

Seurat h5ad Notes
data (normalized) X All genes
counts (raw) raw/X All genes
scale.data (skipped) Can be recomputed with sc.pp.scale()
variable features var['highly_variable'] Boolean column marking ~2,000 genes

Native AnnData Preprocessing

srtdisk now handles “messy” AnnData files natively without requiring Python preprocessing:

  • Column Name Sanitization: Automatically sanitizes obs/var column names during h5ad to h5Seurat conversion
    • Replaces problematic characters (/, spaces, commas, semicolons, colons, backslashes) with underscores
    • Handles duplicate names by appending __dup1, __dup2, etc.
    • Ensures valid R names (prefixes with X if column starts with a number)
    • Example: cell/type -> cell_type, sample name -> sample_name
  • List/Dict Column Conversion: Converts list and dict columns in compound datasets to R-compatible strings
    • Simple lists converted to semicolon-delimited strings (e.g., ["A", "B", "C"] -> "A;B;C")
    • Complex nested structures converted to JSON strings
    • Dict columns converted to JSON strings
  • Nullable Dtype Handling: Properly handles pandas nullable extension dtypes stored as mask+values structures
    • Automatically flattens mask+values groups to simple vectors with NA values
    • Preserves missing value semantics during conversion

Bug Fixes

  • Fixed h5Seurat to h5ad conversion using scale.data (2,000 variable features) as X instead of data (all genes)
  • Fixed handling of obs/var column names containing special characters
  • Improved robustness of compound dataset reading with Python fallback

srtdisk 0.2.0

Release Date: 2026-01-29

Highlights

Full scRNA-seq Conversion Support with Seurat v5

srtdisk now provides complete bidirectional conversion between Seurat objects and h5ad format, with full compatibility for Seurat v5’s new Assay5 architecture:

  • Seurat to h5ad: Use SeuratToH5AD() for direct one-step conversion or the traditional two-step SaveH5Seurat() + Convert() workflow
  • h5ad to Seurat: Use Convert() + LoadH5Seurat() to import scanpy/AnnData objects
  • Properly handles V5 layered data structure (counts, data, scale.data layers)
  • Supports multi-assay objects (e.g., CITE-seq with RNA + ADT)

Metadata Preservation Fixes

  • Fixed categorical metadata loss: Factor/categorical variables in obs are now correctly preserved during h5ad conversion instead of being dropped or converted to strings
  • Improved handling of cell-level and feature-level metadata during round-trip conversions

Spatial Data Support (Visium)

  • Seurat to h5ad: Visium spatial data conversion is fully functional
    • Preserves spatial coordinates, scale factors, and tissue images
    • Compatible with scanpy/squidpy spatial analysis workflows
  • h5ad to Seurat: Spatial conversion fully supported
  • Fixed spatial coordinate X/Y orientation for scanpy compatibility

New Features

  • SeuratToH5AD() wrapper function for convenient direct conversion
  • Enhanced Visium spatial data conversion with proper coordinate handling
  • Support for SlideSeq and FOV-based spatial technologies (experimental)
  • Improved scanpy-compatible naming conventions with standardize = TRUE

Bug Fixes

  • Fixed categorical metadata loss during h5ad conversion
  • Fixed “attribute already exists” error when converting h5ad graphs to h5seurat
  • Fixed GetAssayDataCompat() and SetAssayDataCompat() to use layer parameter (SeuratObject 5.0+ requirement)
  • Fixed spatial coordinate X/Y flip in h5Seurat to h5ad conversion
  • Improved UMAP display in h5ad to Seurat tutorial

Documentation

  • Comprehensive README with quick start conversion examples
  • Comparison table: srtdisk vs SeuratDisk feature improvements
  • Multi-assay (CITE-seq) conversion documentation
  • Spatial data conversion examples (beta)

Known Limitations

  • Multi-assay conversion requires separate h5ad files per assay (AnnData limitation)
  • Large datasets may require sufficient memory for in-memory conversion

srtdisk 0.1.0

New Features

Direct Seurat to H5AD Conversion

  • Added SeuratToH5AD() function for direct conversion from Seurat objects to H5AD format
  • Automatically handles intermediate h5Seurat file creation and cleanup
  • Supports all Convert() parameters including standardize for scanpy-compatible naming

Improved Spatial Data Support

  • Enhanced Visium spatial data conversion between h5Seurat and h5ad formats
  • Better handling of multi-library spatial datasets
  • Improved coordinate transformation and scale factor preservation

Seurat v5 Compatibility

  • Full support for Seurat v5 Assay5 objects
  • Fixed SeuratObject 5.0+ API compatibility (layer vs slot parameters)
  • Proper handling of V5 layered data structure in conversions

Bug Fixes

  • Fixed spatial coordinate X/Y flip in h5Seurat to h5ad conversion (coordinates are now correctly stored as [X, Y] for scanpy/squidpy)
  • Fixed “attribute already exists” error when converting h5ad graphs to h5seurat
  • Fixed GetAssayDataCompat() and SetAssayDataCompat() to use layer parameter (SeuratObject 5.0+ requirement)
  • Fixed vignette assay references (using correct RNA assay instead of non-existent SCT/Spatial)

Code Improvements

  • Removed unused utility functions (WithAssayCompat, GetSlotMapping, GetSeuratSlotMapping, ValidateSlotMapping)
  • Removed deprecated SafeExistsDeprecated function
  • Simplified verbose messaging in AssembleObject.R
  • Extracted CreateFakeCellNames helper to reduce code duplication in Convert.R
  • Removed dead commented-out code

Documentation

  • Updated vignettes with Python/reticulate integration for scanpy/squidpy examples
  • Added conditional evaluation for Python chunks based on package availability
  • Improved vignette examples with bundled test data
  • Added cellxgene spatial dataset download examples

Test Infrastructure

  • Simplified test suite: consolidated 4 test files into single focused test-conversion.R
  • Tests now use real datasets:
    • CellxGene colorectal cancer sample (935 cells) for h5ad testing
    • pbmc3k.final from SeuratData for Seurat testing
    • stxBrain anterior1 from SeuratData for Visium spatial testing
  • Removed synthetic test data in favor of real-world datasets
  • Added UpdateSeuratObject() calls for compatibility with older SeuratData objects

Test Data

  • Replaced pbmc_small.rds with crc_sample.h5ad (CellxGene colorectal cancer, 935 cells x 25,344 genes)
  • Vignettes now use SeuratData (pbmc3k.final, stxBrain) instead of bundled files