Open-source multi-omics

Unified multi-omics in Python. Rust and Torch underneath.

Bulk, single-cell, spatial, and multi-modal — all AnnData-native. Hot loops in Rust; tensor methods on Torch.

$pip install omicverse

Featured projects

The core stack.

/01 omicverse Unified Python library for bulk, single-cell, spatial, and multi-omics analysis. Core · Python /02 anndata-oom Drop-in AnnData replacement that keeps the matrix on disk. Whole pipelines without loading. Data foundation · Rust

Ecosystem coverage

Omics modules already covered by OmicVerse.

One ecosystem spans raw assays, preprocessing, major omics modalities, and domain-specific analysis layers.

DNA Assay FASTA, count matrix, VCF, and alignment-oriented input layers. ov.align

QC Preprocess QC, normalization, PCA, clustering, and batch correction. ov.pp

SC Single-cell Velocity, annotation, NMF, trajectory inference, and cell-state analysis. ov.single

XY Spatial Spatial clustering, deconvolution, neighborhoods, and segmentation workflows. ov.space

BK Bulk Bulk transcriptomics, differential analysis, deconvolution, and pathway context. ov.bulk

FM Foundation Cell embeddings, perturbation models, and foundation-model workflows. ov.fm

3D Molecular Molecular structure, docking, and structure-aware analysis. ov.mol

AT Epigenomics ATAC, ChIP, Hi-C, CUT&Tag, CUT&RUN, and regulatory signals. ov.epi

MB Microbiome 16S workflows, taxonomy summaries, diversity analysis, and abundance testing. ov.micro

MT Metabolomics ID mapping, MSEA, Mummichog, SERRF, biomarkers, and pathway-level readouts. ov.metabol

How it fits together

One Python surface. Rust and Torch in the engine room.

The unified library sits on top of language-appropriate compute backends, with everything speaking AnnData.

/01 One API across data types The same ov.pp / ov.tl / ov.pl surface works for bulk, single-cell, spatial, and multi-modal data — read it once, use it everywhere.

/02 AnnData throughout Inputs and outputs match the scanpy / AnnData conventions, so OmicVerse plugs into pipelines you already have without rewrites.

/03 Rust for hot loops Compute-heavy paths (NMF, Hi-C imputation, BandNorm normalization) are written in Rust — 10–200× speedups with bit-identical numerics.

/04 Torch for tensor methods Tensor and deep-learning methods run on PyTorch with first-class GPU support; classical methods stay on NumPy / SciPy.

From the blog

Recent notes.

All posts

Spatial Analyzing NanoString CosMx Data. Read CosMx data, inspect FOV-aware spatial structure, and prepare downstream spatial modeling. Read → Release What Changed in OmicVerse 2.1.x. Spatial I/O, pure-Python methods, plotting updates, agent runtime work, and new omics modules. Read →