bonsai-rs

Tree representations of single-cell data under Brownian motion. Python bindings for the bonsai-rs crate, a clean-room implementation of Bonsai (de Groot et al., Nature Biotechnology 2026, doi 10.1038/s41587-026-03220-2).

Raw UMI counts in, tree out. Sanity runs first, also in Rust, so the error bars Bonsai needs come from the counts rather than from nowhere.

Install

uv pip install bonsai-rs            # numpy only; dense counts
uv pip install "bonsai-rs[sparse]"  # adds scipy, for sparse counts

Wheels for Python 3.10 and up on Linux x86_64 and macOS.

Use

import bonsai_rs as bs

res = bs.bonsai_from_counts(counts, cell_totals=totals)  # cells x genes

res.tree  # parent array, branch lengths, leaf count
res.node_means  # posterior position of every node, ancestors too
xy = bs.layout(res.tree)  # (n_nodes, 2), edges run node -> parent
newick = bs.to_newick(res.tree, labels=cell_names)

cell_totals is each cell's UMI total over all genes. Leave it out and the row sums are used, which is only right when counts holds every gene.

Already have means and standard deviations? bs.bonsai(means, sds).

Docs

gregorlueg.github.io/bonsai-rs: quickstart, the Sanity handover and its sharp edges, and the API reference.

Development

uv sync
uv run maturin develop --release
uv run pytest
uv run --group docs mkdocs serve

Licence

MIT. Built from the paper and its CC-BY-4.0 Supplementary Information only; see provenance in the repository. Cite the paper.