dendrofan 0.1.1

Circular ("fan") dendrograms for Python, in the visual style of R's ape package

dendrofan

Circular ("fan") dendrograms for Python, in the visual style of R's ape package (plot.phylo(type = "fan"), Paradis & Schliep, 2019) — computed and rendered entirely with NumPy, SciPy, and Matplotlib. No R dependency, no reimplementation of clustering.

Why this exists

scipy.cluster.hierarchy has no circular/fan layout. The usual workaround is a one-off script: call scipy.cluster.hierarchy.dendrogram(..., no_plot=True), take its rectangular icoord/dcoord output, and manually re-project it into polar coordinates with Matplotlib. That works for one figure, but it tends to:

• assume a full 360° circle with no gap for labels at the seam,
• divide by dcoord.max() without checking for a degenerate, all-zero-height tree,
• hardcode SciPy's 10-units-per-leaf spacing convention,
• have no story for trees with 1 or 2 leaves,
• mix clustering, geometry, and plotting into a single script, so none of it is reusable for the next dataset.

dendrofan factors the actually-reusable idea — the rectangular-to-polar coordinate transform — into a small, tested, documented library, and builds a real plotting API on top of it.

Install

pip install -e .          # from a source checkout
pip install -e ".[dev]"   # + pytest, for running the test suite

Quick start

import numpy as np
from dendrofan import circular_dendrogram

rng = np.random.default_rng(0)
data = rng.normal(size=(20, 6))
labels = [f"sample_{i:02d}" for i in range(20)]

result = circular_dendrogram(
    data,
    labels=labels,
    method="ward",       # forwarded to scipy.cluster.hierarchy.linkage
    span=350,             # leave a small angular gap at the seam
    inner_radius=0.15,    # small central gap, ape's fan style
)
result.fig.savefig("tree.png", dpi=200, bbox_inches="tight")

circular_dendrogram accepts the same three kinds of input as SciPy's own dendrogram: raw observations (data=...), a precomputed linkage matrix (Z=...), or a precomputed condensed distance vector (condensed_distances=...).

See examples/quickstart.py and examples/reproduce_station_dendrogram.py (a fully worked, colored-by-group example) for more.

What it handles that the ad hoc version didn't

Case	Ad hoc script	dendrofan
Angular gap at the seam (for label room)	not supported (full circle only)	span=350 (or any value <= 360)
All-merge-heights-equal-zero tree	ZeroDivisionError / silent NaNs	falls back to placing all nodes at the leaf radius
1 or 2 leaves	untested, breaks silently	validated; raises DegenerateTreeError for < 2, works for 2
Mismatched label count	silent misalignment	raises LabelMismatchError
Invalid linkage matrix / distance vector	undefined behavior	raises InvalidLinkageError before plotting
Root at centre vs. rim, radius scale	hardcoded linear, root-at-centre	inner_radius/outer_radius/invert_radius, plus optional "sqrt" area-preserving radius scale
Per-clade coloring	manual, one-off	color_threshold/link_color_func forwarded to SciPy, or a label_colors dict/callable
Scale reference / clade highlighting	not present	dendrofan.annotations.add_scale_ring, highlight_sector
Reuse across datasets	copy-paste the script	one function call

API overview

• dendrofan.circular_dendrogram(...) — the main entry point; draws the tree and returns a CircularDendrogramResult (figure, axes, layout, and the PolarTransform used, for further annotation).
• dendrofan.geometry.PolarTransform — the reusable rectangular-to-polar coordinate map, if you want to bend your own geometry onto the same annulus (e.g. a custom decoration).
• dendrofan.clustering.build_layout(...) — validated wrapper around scipy.cluster.hierarchy.linkage / dendrogram, decoupled from plotting.
• dendrofan.styling.resolve_leaf_colors, legend_handles — color and legend helpers.
• dendrofan.annotations.add_scale_ring, highlight_sector — optional decorations (a distance-reference ring; shaded clade sectors).

Every public function and class has a full docstring; help(...) in a Python session is the fastest way to see the complete parameter reference.

Scope

dendrofan draws circular dendrograms from hierarchical clustering (SciPy linkage matrices) — it does not parse Newick/phylogenetic tree files or handle unequal-tip-depth phylogenies the way ape itself does. If you need that, ape (R) or ete3/Bio.Phylo (Python) are a better fit; dendrofan specifically fills the "I have a SciPy dendrogram and want it circular" gap.

Testing

pytest

The test suite specifically exercises the edge cases listed above (degenerate trees, mismatched labels, invalid linkage matrices, partial spans, single/two-leaf trees).

Citation

If you use dendrofan in a manuscript, please cite the software itself via its Zenodo DOI (see CITATION.cff, or use GitHub's "Cite this repository" button):

Magallanes Quintanar, R. (2026). dendrofan v0.1.0 [Software]. Zenodo. https://doi.org/10.5281/zenodo.21251841

Please also cite the underlying methods it wraps:

• Virtanen, P. et al. (2020). SciPy 1.0: fundamental algorithms for scientific computing in Python. Nature Methods, 17, 261-272. (hierarchical clustering / linkage, which dendrofan builds on)
• Paradis, E. & Schliep, K. (2019). ape 5.0: an environment for modern phylogenetics and evolutionary analyses in R. Bioinformatics, 35(3), 526-528. (for the fan-plot visual convention dendrofan follows)

License

MIT — see LICENSE.