Crate bixverse_sc_data 

The single cell data layer: the bixverse binary sparse format, its streaming readers and writers, and ingestion from h5ad, 10x h5 (feature h5) and MTX.

Feature flags

• analysis (enabled by default) — Plotting helpers and synthetic data generators. Pulls in bixverse-core, and with it faer, SIMD and nearest neighbour search. Switch it off for a reader that only needs the store formats.
• h5 — The h5ad and 10x h5 readers and writers. Pulls in HDF5.
• hdf5-static — Compile the bundled HDF5 C library from source. Leave it off to link an external libhdf5 found via HDF5_DIR or pkg-config, which is what the R package does on Windows: hdf5-metno-src names its output with the host toolchain’s convention, so an msvc host targeting gnu emits libhdf5 and the link fails.
• hdf5-filters — Compile the LZF (32000) and Blosc (32001) filters into the HDF5 build, so h5ad files written by hdf5plugin with those codecs can be read. Both register through H5Zregister and never touch the HDF5 plugin path, which for a hdf5-metno-src build points at a long gone cargo OUT_DIR. Blosc vendors c-blosc plus lz4, zlib and zstd, all built with cc. Snappy is left out on purpose. It is C++, and the R package links this staticlib without the C++ runtime. Filters outside these two (standalone zstd, bitshuffle, lz4, blosc2) raise ScDataErrors::UnsupportedH5Filter. Blosc with snappy passes that check, since the codec sits in the chunk header, and fails in H5Dread with c-blosc naming the missing compressor.
• parquet — Write the h5ad obs / var dataframes and the 10x barcodes and features to Parquet, for DuckDB to ingest with read_parquet. Pulls in parquet without arrow.

Modules

archive_io

Cold-storage archive of the cell-based binary.

bin_merge_io

Helpers for multi-file binary file merging. This is used to combine multiple SingleCells experiments on the R side. It enables the user to combine and merge several experiments.

data_io

Contains the streaming engine for bixverse single cell application. This has the structures for cells/spots and genes ready, plus various utilities, writers and readers.

errors

Errors raised by bixverse-sc-data.

gene_file_io

Builds the gene-based (CSC) binary from the cell-based (CSR) one.

h5_10x_adt_io

Helper functions to specifically extract ADT counts from the h5 file. Can be theoretically used for other modalities, but it does return dense data!

h5_10x_io

10x CellRanger v2/v3 h5 ingestion. Streams the gene-expression modality to the bixverse binarised format. Other modalities (e.g. Antibody Capture) are filtered out here.

h5_10x_multifile_io

Multi-10x loading: scan multiple 10x CellRanger h5 files, apply global gene QC, and write all cells into a single binary file. Mirrors the h5ad and mtx multi-loader designs and reuses the 10x layout helpers from h5_10x_io.

h5_chunked

Parallel reads of gzip-compressed 1-D HDF5 datasets.

h5_filters

Filter pipeline checks for HDF5 inputs.

h5ad_io

Contains the h5ad-related parts to do read in data from h5ad files and transform them into the binarised files for usage in bixverse-rs.

h5ad_multifile_io

Multi-h5ad loading: scan multiple h5ad files, apply global gene QC, and write all cells into a single binary file.

in_memory_io

An in-memory SingleCellReading store over CompressedSparseData2.

mtx_io

Matrix Market (.mtx) ingestion into the bixverse cell store.

mtx_multifile_io

Multi-mtx loading: scan multiple mtx files, apply global gene QC, and write all cells into a single binary file. Mirrors the h5ad multi-loader design.

obs_var_io

Parquet export of the cell and feature metadata: the h5ad obs / var dataframes and the 10x barcodes and features. DuckDB ingests the file with read_parquet, which replaces reading every column into R.

plotting

Helpers to extract genes for plotting. Can densify raw and normalised counts, do Z-score normalisation for groups and caclulate proportion of cells.

prelude

Re-exports for use crate::prelude::*: the sparse prelude (plus the core prelude under analysis) and the single cell data layer.

r_obj_io

Contains the R-based reading in of data (for example for Seurat type files) to transform the counts into the binary Rust-based files.

sc_synthetic_data

Contains helpers for the generation of synthetic data for testing algorithms. Has the option to create single cell-like data with defined cell types, optional batch effects and different cell abundance distributions per given sample, plus a DIALOGUE fixture carrying a planted multicellular programme.

sc_traits

Contains traits and their implementations, specifically designed for the single cell workflows in this crate.

v4_io

The v4 chunk encoding and the v3 -> v4 converters.

Constants

CELL_BATCH_SIZE

How many cells to read in per given batch - global constant

SC_FILE_VERSION

Version of the single cell files writers emit: delta bit-packed chunks with the normalised layer recomputed on read.

SC_FILE_VERSION_V3

Previous file version: absolute u32 indices, normalised layer stored. Readers still open it; v4_io::upgrade_store_to_v4 converts it.

VERSION

Version of this crate.