lucaphylo

name	lucaphylo
description	LucaPhylo — alignment-free phylogenetic inference / placement via hyperbolic protein language models (ESM2 + hyperbolic geometry). Use when: (1) MSA-based trees fail on highly divergent proteins, (2) Exploring PLM phylogeny as a complement to IQ-TREE/PhyloPhlAn, (3) Following the LucaPhylo training or placement checkpoints. Upstream: https://github.com/NickShannn/LucaPhylo. Route via phylogenomics; do not replace GTDB-Tk for MAG taxonomy by default.
license	MIT
category	analysis-tools
tags	["LucaPhylo","PLM","phylogeny","hyperbolic","alignment-free","ESM2"]
upstream	https://github.com/NickShannn/LucaPhylo
stage	taxonomy

LucaPhylo

Upstream: NickShannn/LucaPhylo · Checkpoints: https://huggingface.co/NickShan/LucaPhylo

Citation

Shan, Y. et al. Alignment-free phylogenetic inference via hyperbolic protein language models. bioRxiv https://doi.org/10.64898/2026.05.26.723419 (2026).

See also docs/references.md.

Analytical thinking

Classical phylogenomics needs MSAs. LucaPhylo places unaligned proteins using ESM2 embeddings adapted into hyperbolic space — aimed at highly divergent sequences (training emphasis includes viral RdRP / polyprotein settings). Treat as an AI phylogeny track next to iqtree / phylophlan, not as default MAG taxonomy (gtdbtk).

vs	Prefer LucaPhylo when
iqtree / phylophlan	MSA unreliable or discarded divergent taxa
homology-search	Goal is a tree/placement, not hit tables
gtdbtk	Never as sole bacterial MAG taxonomy

Pin checkpoint stage (1–4), backbone tree, and that results are computational hypotheses. GPU recommended for training; inference benchmarks include CPU/GPU profiles in upstream docs.

How to run

conda create -n lucaphylo python=3.10.16
conda activate lucaphylo
git clone https://github.com/NickShannn/LucaPhylo.git
cd LucaPhylo
pip install -r requirements.txt
# download checkpoints (HF / Zenodo) — see upstream README
# training stages:
#   bash scripts/run_stage_1.sh … run_stage_4_train.sh
# placement / inference: follow Stage-4 inference scripts in upstream docs

Do not invent CLI flags — follow the repository scripts for the installed tag.

Decision tree

Phylogeny without reliable MSA?
├─ Yes, PLM/hyperbolic placement → lucaphylo
├─ Standard MAG/marker phylogenomics → phylophlan → iqtree
├─ Taxonomy placement only → gtdbtk
└─ Homology search only → homology-search

Related skills

phylogenomics · protein-language-model · esm · iqtree · phylophlan · metagenomics-llm · omg · gtdbtk · tool-selection