Release

What Changed in OmicVerse 2.1.x

OmicVerse 2.1.x is not just a small maintenance update. From v2.0.0 to the current development tree, the project accumulated hundreds of commits, hundreds of changed files, and close to one hundred thousand added lines of code. Two major branches, metabolomics and 16S / microbiome analysis, were also moving toward the merge window.

The update focuses on trajectory inference, spatial omics, preprocessing and I/O, plotting, the agent runtime, and new multi-omics modules.

To update:

pip install -U omicverse

Update overview

PP module

• Added a pure-Python pydoubletfinder backend for doublet detection.
• Automatically detects mitochondrial prefixes, avoiding manual MT- versus mt- handling.
• Upgraded Harmony to harmonypy v0.2.0, with GPU, CPU NumPy, and Apple Silicon MLX backends.
• Improved memory handling in torch_pca and covariance_eigh.
• Added the Rust-backed anndataoom out-of-memory AnnData reader.

Single module

• Added ov.single.Monocle, a pure Python rebuild of Monocle 2 that avoids rpy2 and a separate R environment.
• Continued work on dynamic trajectory tools, including lineage-aware trend plotting and improvements around Palantir and Slingshot visualization.

Space module

• Added ov.io.read_xenium for full Xenium outs/ ingestion.
• Converts Xenium cell_boundaries into WKT polygons that can be passed directly to ov.pl.spatialseg.
• Added ov.io.write_visium_hd_cellseg to export cell-level AnnData back into a Space Ranger v4-compatible directory structure.
• Added CosMx FOV-aware plotting, including multi-FOV layout and per-FOV overlays.
• Added CellSAM support and renamed stardist() to the more general cellseg().
• Added method="cellcharter" to ov.utils.cluster.

Plotting module

• Added Marsilea-based heatmap plotting.
• Added a set of cell-cell communication plotting APIs.
• Added ov.pl.create_custom_colormap.
• Added half-violin boxplots and started consolidating older plotting APIs.
• Improved legend behavior for subset plotting.

Agent module

• Reworked ov.Agent, Jarvis, and OVAgent runtime internals.
• Added Codex OAuth support.
• Moved multiple message channels onto a unified MessageRuntime.
• Added sandboxing, timeouts, stdout guards, retries, and security hardening.

New modules

• Added ov.metabol for metabolomics workflows, including ID mapping, MSEA, Mummichog, SERRF, DGCA, ASCA / MixedLM, ROC, and biomarker panels.
• Added ov.micro and ov.alignment for 16S / microbiome analysis from amplicons to microbiome AnnData objects.

Why the update matters

The theme of this release is reducing dependency friction while making high-resolution spatial workflows more complete. Xenium, Visium HD, and CosMx are no longer only file formats that can be read; they now have fuller analysis paths that connect I/O, preprocessing, segmentation, plotting, and export.

The pure-Python Monocle 2 rebuild points in the same direction. When core methods no longer require an R bridge, they become easier to install, easier to run in notebooks and pipelines, and easier to expose to agents.

For day-to-day users, the headline is practical: update OmicVerse if you need modern spatial data support, more robust preprocessing, richer plotting, or a more capable agent runtime.