Why it exists

Annotating a tree for a paper usually means a spreadsheet: a column of leaf names copied out of the tree, a column of categories typed next to them, saved as a template, uploaded to a web service, and repeated every time a name is wrong or a clade moves. Phylo3s was built for one lab that was tired of that loop. The difference is structural: the app already knows every leaf name, so the annotation table is made in the app, by right-clicking a clade, by filtering, or a cell at a time. Two actions annotated 309 of the 313 leaves in the reference tree.

It runs two ways from one codebase: as an offline desktop application installed from a single file, and as a self-hosted web app on a server. The figure is drawn by the same code in both.

Phylo3s showing a 313-leaf tree in rectangular layout with two colour strips, a presence/absence matrix and clade ranges, a legend on the right, and the layout panel — **fig. 01** — a 313-leaf tree with four annotation tracks: two colour strips, a presence/absence matrix and clade ranges. Everything on the right is drawn beside the figure and exported with it.

What it does


upload	Newick, Nexus, PhyloXML and jplace, with MrBayes and NHX metadata; parsed once, server-side
render	rectangular, circular and unrooted; phylogram or cladogram; interactive at 10,000 leaves
interact	pan, zoom to cursor, click to select, right-click for root-here, collapse, prune; leaf search with highlighting
topology	re-root anywhere, midpoint root, ladderize, restore; 25 steps of undo, annotation edits included
annotate	colour strips, presence/absence matrices, clade ranges, heatmaps, bar charts; built in the app or loaded from a table
interoperate	iTOL `DATASET_*` templates load and save for all five kinds; a dataset written out and loaded back draws the identical figure
support values	a dot sized across a chosen range, or the number itself; detected as bootstrap or posterior probability and named so in the legend
export	SVG, PNG at 300 dpi, vector PDF, all from the same command list the screen uses
keep	saved state per tree, named views, projects and workspaces

The same tree in circular layout, with the clade ranges as tinted sectors and the colour strips as outer rings — **fig. 02** — circular. Ranges become sectors, strips become rings; arc, rotation and inner radius are sliders.

Annotation, made in the app

This is the part that sets the tool apart. Annotation lives in the left rail under the trees, because it is a subject of the tree rather than a control for the figure. The table itself opens over the canvas, where there is room for thirty rows at once.

The annotation table open over the canvas: a dataset named Phyla with two categories, a grid of leaf names with their values, and filter fields — **fig. 03** — the table, made here. Name a dataset, declare the categories a leaf can carry, then fill them in three ways: right-click a clade on the tree, filter the grid and press a category chip, or type a cell at a time. No spreadsheet, nothing to upload; the app already knows every leaf name.

The Style panel: leaf label options, leader colour swatches, font family and size, branch width, and the bootstrap section — **fig. 04** — style. Every slider has a number box that accepts values beyond its range.

The project page: a table of trees with leaf counts, dataset colour chips, dates, and upload and delete buttons — **fig. 05** — projects. A project holding trees refuses to be deleted; deleting a tree says what goes with it.

What it exports

The exported figure: the rectangular tree with support dots, four annotation tracks and a legend listing Phyla, Domain, Motif, type and Bootstrap — **fig. 06** — the PNG export of fig. 01, 15,017 × 11,884 px at 300 dpi, reduced here. SVG and PDF are the same figure in vector form.

one figure, three files

fig. 07 — the PDF is converted server-side from exactly the SVG the SVG button writes, so the two cannot differ

Install

Each installer carries its own Python, API, browser engine, interface and print typefaces. The machine needs no Python, no Node and no internet.

ubuntu · one command

fig. 08 — Windows runs the setup file; any Linux runs the AppImage; the same codebase serves a group as a web app

v0.1.1

fig. 09 — three installers, one codebase, every check scripted in headless chromium

Design decisions

The table is made in the app. That is the whole reason this exists rather than being a skin over iTOL.
One command list. Screen, SVG, PNG and PDF are the same figure, so what you see is what prints.
Undo covers everything. Re-rooting, ladderizing and marking a clade share one stack; the button’s tooltip names the step it will take back.
Nothing fetched. The interface font is vendored; a check asserts every request the page makes is same-origin.

What is not done

No macOS build yet: it needs a Mac to build on. DATASET_TEXT, DATASET_PIECHART and the iTOL label, collapse and prune files are refused by name, because a kind the renderer cannot draw must not be stored.