v1.2.0

Multi-column grouping

Repeat --column / -c in build or col to group cells by a combination of metadata values:

refmat col sample.rds -c condition -c celltype
refmat build sample.rds -c condition -c celltype -o combined.tsv

This produces group labels such as wt_macrophage and mut_astrocyte. Use --separator ':' to customize the separator. Supported for Seurat, SingleCellExperiment, and AnnData.

• Column argument order determines label order.
• Combined groups contain observed combinations only, in first-appearance order.
• Cells missing any selected grouping value are excluded and reported on stderr.
• Ambiguous joined labels cause an error rather than silently merging distinct groups.
• Single-column syntax and ordering remain supported.

All 17 local tests pass, including combined group means across the three formats. GitHub Actions tests and builds Apple Silicon macOS and AMD64 Linux binaries; SHA256SUMS accompanies the archives.

v1.1.2

Fixes

• Preserve metadata row names when reading RDS files lazily through the pinned rds2rust fork. This fixes false "metadata has 0 cells" errors for Seurat objects whose first metadata column is a factor, while retaining cell-order validation.
• Keep large expression and metadata vectors lazy.

Downloads

• refmat-v1.1.2-aarch64-apple-darwin.tar.gz: Apple Silicon macOS.
• refmat-v1.1.2-x86_64-unknown-linux-gnu.tar.gz: AMD64 Linux with glibc 2.35 or newer.
• SHA256SUMS: checksums for both archives.

Extract the matching archive and put refmat on your PATH. HDF5, gzip/LZF, and the RDS xz decoder are built into the binary; R and Python are not required to run it.

Validation

GitHub Actions tests and builds both platforms before publishing release assets. The affected Hayashi dataset also passed local validation: 22,639 features by 10 groups, matching an independent R calculation within 8.9e-16.

refmat v1.1.1

New features:

• Seurat v3 and v4 object support
• consistency of column arguments between col and build commands

refmat v1.0.0

First stable release of refmat.

Features

• Read Seurat v5, SingleCellExperiment, and AnnData objects.
• Preview cell metadata with refmat head.
• Count cells by metadata value with refmat col.
• Build genes-by-cell-type reference matrices with refmat build.
• No R or Python required at runtime.
  See the README for installation and usage instructions.