OpenRelTime

Language: English · 中文

OpenRelTime is an open-source Python implementation of the relative rate framework (RRF) for non-Bayesian molecular dating (Tamura et al. 2018, MBE 35:1770-1782), the method behind RelTime in MEGA. It estimates relative lineage rates and divergence times from a branch-length tree, converts calibrations (min/max bounds or probability densities) into absolute times with analytical confidence intervals, and adds CorrTest rate-autocorrelation testing, birth-death (ddBD) tree-prior estimation and a bag-of-little-bootstraps (BLB) pipeline — everything scriptable from Python or the command line.

• Documentation (English): docs/README.md — user guide, parameter handbook, methods, validation, tutorials
• 中文文档（全手册亦有中文版）：README-zh.md · 文档目录 · 使用说明
• API/CLI equivalence, reproducible golden standards, GPL-3.0

@software{openreltime,
  author    = {曾, 子超},
  title     = {OpenRelTime: relative rate framework based molecular dating in Python},
  year      = {2026},
  version   = {v0.1.1},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.23053145},
  url       = {https://github.com/ZengZichao/OpenRelTime}
}

Why OpenRelTime

OpenRelTime reproduces R3F numerically on the 274-tip mammalian benchmark (regression slope 1.0000; observed median relative deviation 6.8e-16 for rates and 8.7e-16 for times). The automated gate G2a/G2b asserts a slope >= 0.999 and a median relative deviation <= 1e-6; the largest observed deviation (~4.8e-15 rates, 3.6e-14 times) is measured but not asserted by the gate — see docs/validation.md. On top of the shared core OpenRelTime adds the modules R3F lacks — calibration conversion and analytical confidence intervals (msz236) — plus an end-to-end BLB pipeline.

Installation

Distribution. This release ships as source plus the sdist and wheel attached to the v0.1.1 release. It is not on a package index yet; an index release will be announced here.

From a checkout:

git clone https://github.com/ZengZichao/OpenRelTime
cd OpenRelTime
pip install --editable ".[dev,plot]"   # core (numpy, scipy, pandas, click)
                                       # + matplotlib visualisation ([plot])
                                       # + pytest/ruff/mypy ([dev], contributors)

Or install the released wheel directly from GitHub — no clone, and the [plot] extra brings matplotlib:

pip install "openreltime[plot] @ https://github.com/ZengZichao/OpenRelTime/releases/download/v0.1.1/openreltime-0.1.1-py3-none-any.whl"

The desktop GUI is a separate product, installed separately (see below).

plot and dev are the only extras the package declares; combine them in one call, e.g. ".[plot,dev]".

Requirements: Python >= 3.10, Linux/macOS/Windows. There are no compiled extensions; installation takes seconds. CI runs the test suite on Linux (ubuntu-latest) across Python 3.10, 3.11, 3.12 and 3.13, and on macOS (macos-latest) on Python 3.13 only — the three older macOS combinations are excluded from the matrix in .github/workflows/ci.yml to keep the runner queue short, so macOS is a single-interpreter check rather than a 3.10-3.13 claim. Windows (windows-latest) runs both the engine and the end-to-end suites in a separate job on Python 3.13; the job was added for 0.1.1, so its first evidence is the next Actions run (see docs/validation.md for the tested-vs-claimed breakdown and for the one Windows limitation the suites do not cover).

OpenRelTime Studio (desktop GUI) — separate product

OpenRelTime Studio is a native desktop app for researchers who prefer point-and-click: load a Newick/NEXUS tree, run RRF, add calibrations by clicking internal nodes on the canvas, then run calibration, analytical CIs, CorrTest and ddBD — all through this engine's public Python API, with CSV/NEXUS/JSON/PNG export and a reproducible equivalent-command-line panel. Its interface is fully bilingual (English / 简体中文) and switches live, it ships light and dark themes, a hand-drawn vector SVG icon, and a built-in example tree with demo calibrations so first-time users need no data.

It is not part of this repository and is not installed from it. It ships as its own project and depends on the engine as an ordinary third-party package. Neither is on an index yet, so install both wheels from their releases:

pip install \
  "openreltime[plot] @ https://github.com/ZengZichao/OpenRelTime/releases/download/v0.1.1/openreltime-0.1.1-py3-none-any.whl" \
  "OpenRelTime-Studio @ https://github.com/ZengZichao/OpenRelTime-Studio/releases/download/v0.1.1/openreltime_studio-0.1.1-py3-none-any.whl"
openreltime-studio

On Apple Silicon, Studio's release also attaches a prebuilt OpenRelTimeStudio-v0.1.1-macOS-arm64.zip that needs no Python at all.

Home page: https://github.com/ZengZichao/OpenRelTime-Studio · GUI manual: see that project's documentation. Installing the engine alone leaves the Python API and every CLI subcommand above unchanged — Studio adds a front end, never new analysis code.

Quick start (30 seconds)

import openreltime as ort

tree = ort.read_tree("example.nwk", outgroup=["Out1", "Out2"])
times = ort.rrf_rates_times(tree)             # rates + relative times
times.write("quickstart", with_rate=True, nexus=True)   # CSV + NEXUS timetree

Equivalent CLI:

openreltime rates-times -i example.nwk --outgroup "Out1,Out2" -o quickstart

From an alignment (BLB, requires IQ-TREE):

openreltime blb -a alignment.fasta --iqtree iqtree --seed 42 -o blb_out

Features

• RRF core — geometric mean (R3F/msy044-compatible) and arithmetic mean (PNAS 2012) conventions; extreme-rate guard; polytomy handling; Newick + NEXUS ([&rate=...]) I/O.
• Calibration — hard min/max bounds and probability densities (uniform/exponential/normal/lognormal) converted to absolute times via the global factor f; the effective-bounds resampling method of msz236.
• Analytical CIs — delta method (msz236 eqs 7-14) with flexible sampling variance sources and truncation at hard bounds; empirical node-age CIs from the effective-bounds replicates.
• CorrTest — autocorrelated-rates test with the published fixed-coefficient logistic model.
• ddBD — birth-death speciation prior (birth, death, sampling fraction) for Bayesian dating pipelines.
• BLB pipeline — two-level site resampling through IQ-TREE with per-clade time aggregation.
• MEGA-CC bridge (optional) — generate .mao files, run megacc, parse RelTime output back to OpenRelTime tables.

Command-line overview

openreltime rates        -i tree.nwk [--outgroup og.txt] [--mean geometric] [-o prefix]
openreltime times        -i tree.nwk [--normalize] [--no-guard] [--r3f-compat] [-o prefix]
openreltime rates-times  -i tree.nwk [--plot timetree.png] [--r3f-compat] [-o prefix]
openreltime calibrate    -i tree.nwk -c calibrations.tsv [--method effective] [-o prefix]
openreltime ci           -c <calibrated prefix> [--n-sites 1000] [-o prefix]
openreltime corrtest     -i tree.nwk [--sister-resample 100] [-o prefix]
openreltime ddbd         -i tree.nwk [--anchor-time 1.85] [-o prefix]
openreltime tree2table   -i tree.nwk [--time] [-o prefix]
openreltime monophyly   -i tree.nwk [--taxon-table taxa.tsv] [--groups A,B] [-o prefix]
openreltime blb          -a aln.fasta --iqtree iqtree [--gamma 0.7 ...] [-o dir]
openreltime megacc       -i tree.nwk -c calibrations.tsv --megacc /path/to/megacc

Every analysis stage whose parameters can move the numbers (rates, times, rates-times, calibrate, ci, corrtest, ddbd) merges its parameters, warnings and random seed into <prefix>_report.json; runs that share a prefix contribute their own key. tree2table, monophyly, blb and megacc write only their table.

Calibration file (calibrations.tsv)

node_id	taxon_set	min_bound	max_bound	density	density_params
	Homo_sapiens|Pan_troglodytes	6.0	8.5	.
	Elephas_maximus|Loxodonta_africana	.	.	exponential	offset=60;mean=20
	Dasypus_novemcinctus|Choloepus_didactylus	70	105	.

Identify nodes by MRCA of a taxon set or by internal node id; give hard bounds, a density, or both. Densities: uniform, exponential, normal, lognormal (parameters as key=value;...).

The three rows above show the syntax only. As numbers they are placeholders that no rate rescaling of a single tree can satisfy jointly, and calibrate now stops there and names the offending nodes instead of returning the broken ages; a density without a hard bound is likewise refused under the default --method bounds. For a table that runs as it stands, use data/examples/example_calibrations_feasible.tsv; both rules are walked through in docs/tutorials/tutorial2-end-to-end.md.

Validation

The test suite encodes the project's acceptance gates; each gate, what it actually asserts, and which shipped modules it does not cover are listed in docs/validation.md:

python -m pytest tests/

• G1 hand-computed 3-/4-taxon RRF values (both conventions, 1e-12)
• G2a/G2b R3F regression on the 274-tip mammalian tree (gate asserts slope

  = 0.999 and median relative deviation <= 1e-6; observed medians 6.8e-16 rates / 8.7e-16 times)

• G2c NEXUS [&rate=...] interop round-trip
• G6/G7 CorrTest agreement with the stored R3F CorrScore (0.9996); calibration bounds/conflict handling
• Golden files + example inputs + the R script that produced the goldens ship in data/golden/r3f/, data/examples/ (see its README.md), data/PROVENANCE.md and reproduce/golden_r3f.R.

Documentation

The full manual set exists in both English and Chinese; each document links to its counterpart at the top. Index: docs/README.md.

License and reuse

GPL-3.0-or-later (LICENSE). Behavioural references (R3F, MEGA sources, ape) are credited in THIRD-PARTY-NOTICES.md with per-file provenance in data/PROVENANCE.md. The CorrTest repository (no licence) was not copied; only published constants of Tao et al. (2019) are used.

Authors

曾子超 (Zichao Zeng) · zengzichao@sjtu.edu.cn · ORCID 0000-0001-6553-970X

The preferred citation is recorded in CITATION.cff. Cite 10.5281/zenodo.23053145, the engine's Zenodo concept DOI, for the software across versions; 10.5281/zenodo.23053146 is the archived v0.1.0 version record, and the matching version record for 0.1.1 is minted when the 0.1.1 release is cut.