PIASO

Precise Integrative Analysis of Single-cell Omics

PIASO is a Python and Rust toolkit for single-cell omics (scRNA-seq, scATAC-seq and spatial transcriptomics), covering the analysis from raw counts through to the figures in a paper. Performance-critical routines are implemented in Rust.

Installation

Install from PyPI (stable release):

pip install piaso-tools

PIASO needs Python 3.10 or later. Pre-compiled wheels are built for Linux on x86_64, CPython 3.10, 3.11 and 3.12, and there pip install has nothing to build. On any other platform or Python (macOS, Windows, Linux on ARM, Python 3.13 or later), pip builds PIASO from its source distribution, which needs a Rust toolchain and a C compiler; without them the installation stops with a build error.

This also installs cytome, the on-disk dataset format PIASO reads and writes. Nothing extra to install to work with .cytome files.

Install from GitHub (latest development version; builds from source, so it needs Rust on every platform):

pip install git+https://github.com/genecell/PIASO.git

The bioconda package (conda install -c bioconda piaso) is version 1.0.3, from before the Rust extension; install from PyPI for the current release.

Documentation

piaso.org: tutorials, API reference and release notes.

Using PIASO with a coding agent

PIASO-for-agents makes the PIASO ecosystem available to coding agents from one canonical knowledge base, generating Claude skills, Cursor rules, AGENTS.md, llms.txt, and an MCP server. Useful if you work in Claude Code, Cursor, Copilot, Codex, Windsurf, Cline, or Aider and want the agent to know the current API rather than guess it.

Any model with web access can be pointed straight at:

https://piaso.org/llms.txt
https://piaso.org/llms-full.txt

A first analysis

import piaso, cosg

piaso.tl.infog(adata, n_top_genes=3000)                       # normalize + select
piaso.tl.runSVD(adata, layer="infog", n_components=50, key_added="X_svd")
piaso.tl.neighbors(adata, use_rep="X_svd"); piaso.tl.leiden(adata)
piaso.tl.umap(adata, use_rep="X_svd")
cosg.cosg(adata, groupby="leiden")                            # marker genes
piaso.pl.embedding(adata, color="leiden")

That whole workflow runs on a plain pip install piaso-tools, with no scanpy required.

Who this is for

• You have a single-cell dataset and want an analysis, not a toolchain. Reading, QC, normalization, dimensionality reduction, clustering, marker genes, annotation and plotting are one package with one set of conventions.
• Your data outgrew memory. The same function calls run on an AnnData in RAM or stream from a cytome file on disk, where peak memory is set by the batch size instead of the cell count. Validated to several million cells.
• You care what the figure looks like. The plotting suite and piaso.settings are built for publication figures rather than for quick looks.
• You work with a coding agent. The API is published in an agent-readable form (see above), so the agent works from the current signatures.

If you only need one method, the pieces are usable on their own: cosg for markers, cytome for storage.

The ecosystem

PIASO is the analysis layer of a small set of packages that fit together, and each is useful alone:

Contributing

Issues and pull requests are welcome at github.com/genecell/PIASO. Bug reports are most useful with the output of piaso.__version__ and a minimal example.

Citation

If PIASO is useful for your research, please consider citing Wu, S.J., Dai, M. et al. Pyramidal neurons proportionately alter cortical interneuron subtypes. Nature (2026). https://doi.org/10.1038/s41586-025-09996-8

Contact

Min Dai dai@broadinstitute.org