AI for Bio · Model

BioReason

Multimodal model feeding a DNA foundation model into an LLM for pathway reasoning and variant effects

Genomics & RNAAcademic labCanadaSuperseded by a successor中文

Overview

NameBioReason · aliases: BioReason
Full nameNo separate expanded name
Ambiguous names (not auto-linked)None
FamilyNot a family entry. Single-model entry, not a family
OrganizationUniversity of Toronto, Vector Institute, University Health Network
Organization typeAcademic lab
CountryCanada
Model classDNA-language multimodal reasoning
DomainGenomics & RNA (also: Text, LLMs & agents)
First release · preprint
Basis: arXiv 2505.23579 v1 submitted source
Other milestones
  • Preprint v1: · arXiv v1 source
Latest update · successor
Basis: bioRxiv v2 of the successor BioReason-Pro
Successor BioReason-Pro: bioRxiv v1 2026-03-20, v2 2026-07-19; BioReason itself has arXiv v2 and NeurIPS 2025 proceedings source
The most recent sourced event, not a claim that nothing changed since.
StatusSuperseded by a successor · Successor BioReason-Pro exists
OpennessCode: Open
Weights: Open
Code on GitHub, datasets on Hugging Face
Benchmarks reported1 counted (modified protocol 1)
Wet-lab validationNone recorded
Unverified fieldsNone
As of2026-11-16

What it does

Feeds representations from a DNA foundation model (Evo 2 or Nucleotide Transformer) into a Qwen3 language model for KEGG pathway reasoning and variant-effect prediction, with reasoning given as text.

Benchmarks reported

Every row is one pairing from the survey's relation file; a score belongs to the benchmark page's ranking, where it is shown with metric, eval set, source type and date. 1 of 1 pairings are counted.

BenchmarkRelationEval set / protocolSource typeNote and evidence
11 ClinVar hold-out protocolDeveloper report, modified protocol—Self-reportedIts own pathway question task, not this protocol · evidence

Other evaluations (outside the 44)

None recorded.

Developers report: Source states no developer claim

Independent evidence: No independent evidence found. No independent evidence found

Training data

Group in Figure 7: Public data only.

ClassValueDetail
a. Public databasesYesKEGG pathways plus ClinVar, OMIM, dbSNP and COSMIC annotations; gnomAD benign controls
b. Private or licensed dataNot disclosed
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: Academic project of the University of Toronto, Vector Institute and University Health Network; accepted at NeurIPS 2025
  • Funding: Not disclosed or not applicable
  • Products and partnerships: No product or partnership found
  • Related models: Evo 2, scGPT / Geneformer

Reading notes

  1. Its only pairing is a self-built task, not comparable with the official ClinVar protocol
  2. 2026 activity comes from the successor BioReason-Pro, not a new BioReason version

Caveats

  1. The BioReason arXiv v2 date (2025-10-24) was not re-verified
  2. No wet-lab validation

Sources

  1. arXiv preprint 2505.23579 (preprint)
  2. GitHub repository bowang-lab/BioReason-Pro (repository)
  3. GitHub repository bowang-lab/BioReason (repository)
  4. Journal article DOI 10.52202/085713-3728 (journal)

Corrections

2 corrections
FieldWasNowBasisFound
latest_update2026-08-20 (successor repository commit)2026-07-19 (BioReason-Pro bioRxiv v2)source2026-11-16
latest_update.descriptionSuccessor repository creation and commits described as the updateChanged to the successor preprint events; repository commits are not releasessource2026-11-16

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