Bio Infra for Future AI

Open-source, open AI infrastructure, built together — OmicVerse.

Browse packages R-port catalog

omicverse/omicverse

1.4kstars 150 forks 20 open issues GPL-3.0

Install $pip install omicverse Installation guide ›

R ports

R packages, rebuilt function by function.

Every port maps the upstream R functions to Python or Rust and is checked against the R reference on a fixed fixture. Click a sticker to see its map.

70packages

1,490function mappings

84%ported

Open the catalog How we port

Benchmark

1.4M cells in about a minute, from disk.

Full scRNA-seq workflow on SEA-AD: 67 s and 13.5 GB of memory.

Runtime logPeak memory

CPU · 64 threads

omicverse[cpu]67 s13.5 GB

brisc61 s114 GB

Scanpy*1.4 h256 GB

Seurat + BPCells*2.9 h15 GB

Seurat v6> 7.8 h crashed288 GB

GPU · 1× H100 + 16 threads

omicverse[gpu]71 s5.2 GB

omicverse[mixed]2.6 min13.4 GB

rapids-singlecell*3.8 minn/a

SEA-AD · 1.38M cells × 36,601 genes · * brisc paper: 192-core node; rapids-singlecell on 4× H100

Ecosystem

Every omics layer, one import.

Ten modules share one AnnData-first API. scanpy pipelines plug in unchanged.

import omicverse as ov

Metabolomicsov.metabol

Quickstart

Count matrix to clusters in eight lines.

The object stays AnnData the whole way, so scanpy, exports or an agent can take over at any step.

$ pip install omicverse

import omicverse as ov
adata = ov.datasets.pbmc8k()
ov.pp.qc(adata, tresh={"mito_perc": 0.2, "nUMIs": 500})
ov.pp.preprocess(adata, mode="shiftlog|pearson", n_HVGs=2000)
ov.pp.scale(adata); ov.pp.pca(adata, layer="scaled")
ov.pp.neighbors(adata, use_rep="scaled|original|X_pca")
ov.pp.umap(adata); ov.pp.leiden(adata)
# adata.obs["leiden"] · adata.obsm["X_umap"]