AI for Bio · Model

LucaOne

Unified pretrained foundation model for DNA, RNA and protein sequences, about 1.8B parameters

ProteinsMixedChinaActive中文

Overview

NameLucaOne · aliases: LucaOne
Full nameNo separate expanded name
Ambiguous names (not auto-linked)LucaCell, LucaVirus
FamilyNot a family entry. Single-model entry, not a family
OrganizationAlibaba Cloud, Sun Yat-sen University
Organization typeMixed · Joint work by a cloud business unit and a university
CountryChina
Model classUnified nucleic-acid and protein model
DomainProteins (also: Genomics & RNA)
First release · preprint
Basis: bioRxiv v1 preprint (repository creation date not used) source
Other milestones
  • Preprint v1: · bioRxiv v1 source
  • Journal online: † · Journal online publication (per the earlier record; Crossref fetch failed) source
  • Weights release: · LucaOne weights uploaded to Hugging Face source
Latest update · successor
Basis: Successor LucaCell weights added on Hugging Face
Successor LucaCell weights released; LucaVirus came earlier in 2025 and the last GitHub commit is bug fixes source
The most recent sourced event, not a claim that nothing changed since.
StatusActive · Successor weights released within 12 months
OpennessCode: Open · Apache-2.0
Weights: Open · Apache-2.0
Code and weights Apache-2.0; the pretraining corpus is archived in a public genome database
Benchmarks reported0 counted · none of the 44
Wet-lab validationNone recorded
Unverified fields† release_milestones.journal_online
As of2026-11-16

What it does

Unified semi-supervised multi-task pretraining over DNA, RNA and protein sequences at about 1.8B parameters, with a corpus spanning 169,861 species; downstream tasks use simple heads.

Benchmarks reported

None of the 44 benchmarks. This model is on the zero-pairing list.

How it evaluates itself: 8 downstream validation tasks; Central-dogma few-shot analysis

Other evaluations (outside the 44)

  • Eight downstream validation tasks (Own task set) Self-reported: Author-chosen DNA, RNA, protein and mixed-input tasks compared with several reference models
  • Central-dogma few-shot analysis (Case study) Self-reported: Shows the DNA-protein correspondence the model learned in few-shot settings

Developers report: Developers report a unified nucleic-acid and protein foundation model over 169,861 species that captures central-dogma relations and is competitive on eight downstream tasks

Independent evidence: Two third-party evaluations (2026) split: strong under supervised random splits but clearly weak zero-shot; viral-taxonomy results vary by split, and one listing marks its numbers as self-reported

Training data

Group in Figure 7: Public data only.

ClassValueDetail
a. Public databasesYesDNA, RNA and protein sequences from 169,861 species archived in a public genome database; total counts not confirmed in the paper
b. Private or licensed dataNo
c. Own wet-lab data used for trainingNo
d. Wet lab used only for validationNo

Wet-lab validation

None. The developers report no wet-lab validation.

Background

  • Organization: Published jointly by Alibaba Cloud (Hangzhou) and Sun Yat-sen University (Shenzhen); no separate funding
  • Funding: Not disclosed or not applicable
  • Products and partnerships: The same team also built a virus-discovery model and the successors LucaVirus and LucaCell
  • Related models: Evo 2, NTv3, xTrimoPGLM

Reading notes

  1. All self-evaluations are bespoke and outside the 44 benchmarks; on the zero-pairing list under this round's rules
  2. Third-party work shows a large gap between supervised and zero-shot settings; do not cite only one side

Caveats

  1. The journal online date comes from the earlier record and was not re-checked via Crossref
  2. Total corpus counts are not confirmed in the paper

Sources

  1. Hugging Face page api/models?author=LucaGroup (repository)
  2. bioRxiv preprint (preprint)
  3. GitHub repository LucaOne/LucaOne (repository)
  4. Journal article DOI 10.1038/s42256-025-01044-4 (journal)
  5. Third-party nucleic-acid model benchmark (third-party paper)
  6. Third-party viral genomics benchmark (third-party paper)

Corrections

None recorded.

Part of the AI for Bio benchmarks survey · all 112 models · open in the explorer. Record as of 2026-11-16.