AI for Bio · Model

Quine

Microsoft Research's biology world model plus research harness, research use only

Text, LLMs & agentsBig-tech labUnited StatesProduct only, no public artifact中文

Overview

NameQuine · aliases: Microsoft Quine
Full nameA product name with no expanded form
Ambiguous names (not auto-linked)Quine
FamilyNot a family entry. A single system with no later version
OrganizationMicrosoft Research
Organization typeBig-tech lab
CountryUnited States
Model classAgent / world model + harness
DomainText, LLMs & agents (also: Single-cell & perturbation)
First release · product update
Basis: Microsoft Research official blog source
Other milestones
  • First public evidence: · Official blog source
  • Official announcement: · Official blog source
Latest update · no later event
Basis: No update found after launch
No later sourced event found (as of 2026-11-16); access is limited to Quine Fellows and selected collaborations, with broader access via Microsoft Discovery planned without a timeline source
The most recent sourced event, not a claim that nothing changed since.
StatusProduct only, no public artifact · Closed, restricted access
OpennessCode: Closed
Weights: Closed
Initially limited to the Quine Fellows programme and selected collaborations, research use only
Benchmarks reported0 counted (unconfirmed 6)
Wet-lab validation1 recorded campaign
Unverified fields† wetlab_validation[0], training_data.c_own_wetlab_for_training
As of2026-11-17

What it does

Two parts: a multimodal world model spanning genomics, proteins, chemistry, RNA, cell state and bioimaging, and a harness linking models, scientific tools, literature and researchers; Quine makes predictions that are tested in the wet lab. Compared with hypothesis-focused Co-Scientist and tool-orchestrating Biomni, it ties its own predictive models to a wet-lab partner within one research programme; for cell-state prediction compare STATE.

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. 0 of 6 pairings are counted.

BenchmarkRelationEval set / protocolSource typeNote and evidence
14 PubMedQAUnconfirmed (not counted)—Unconfirmed sourceNo report found in official materials †disputed · evidence
18 LAB-Bench / LABBench2Unconfirmed (not counted)—Unconfirmed sourceNo report found in official materials †disputed · evidence
23 BioASQUnconfirmed (not counted)—Unconfirmed sourceNo report by it was found in the official material. †disputed · evidence
29 BixBenchUnconfirmed (not counted)—Unconfirmed sourceNo report found in official materials †disputed · evidence
41 BioML-benchUnconfirmed (not counted)—Unconfirmed sourceNo report on this benchmark found in official materials. †disputed · evidence
42 BioMysteryBenchUnconfirmed (not counted)—Unconfirmed sourceNo report on this benchmark found in official materials. †disputed · evidence

Other evaluations (outside the 44)

  • Site text and agent benchmarks (none disclosed) Self-reported: Checked LAB-Bench, BixBench, BioASQ, PubMedQA, BioML-bench, BioMysteryBench: the official blog gives no benchmark score
  • PDAC cell-state transition prediction (Wet-lab validation) partner: With Broad, compound rankings tested on patient-derived ex vivo models; counts not disclosed

Developers report: a multimodal biology world model plus research harness narrows candidates quickly (about a weekend), and its top predictions were borne out in wet-lab assays.

Independent evidence: No independent evidence found. Only press commentary; no third-party reproduction or quantitative independent evaluation found

Training data

Group in Figure 7: Public data + wet lab for validation only.

ClassValueDetail
a. Public databasesNot disclosedNo datasets named source
b. Private or licensed dataNot disclosed source
c. Own wet-lab data used for trainingNot disclosedPancreatic-cancer patient-derived ex vivo models come from a multi-year collaboration with Broad; whether they were used for training is not stated source
d. Wet lab used only for validationYesPancreatic-cancer cell-state predictions were tested in ex vivo experiments on the Broad side source

Wet-lab validation

Every rate carries its numerator and denominator. Claims that could not be verified are shown as unverified, without numbers; different paper versions are separate entries.

PDAC cell-state transition (developer blog)
Targets — · Designs tested — · Hits — · Success — · Best affinity —

Denominator basis: No counts disclosed: the blog says only that "thousands" of compounds were ranked and "a handful" tested
The blog says top-ranked compounds gave the largest intended state shifts and others were predicted and confirmed to push cells to a third phenotype; no hit rate, negatives or controls
Executor: Ex vivo experiments at partner Broad Institute; not independent replication

Self-reported · · †unverified · source

Background

  • Organization: Microsoft Research is Microsoft's research division, with areas such as AI for Science and Health Futures and labs in Redmond, Cambridge (UK and US), New York, Montreal, Bangalore, Beijing, Nairobi and elsewhere.
  • Funding: Microsoft's internal research budget; no separate funding.
  • Products and partnerships: Project Ex Vivo is a joint Microsoft and Broad Institute cancer research collaboration supported by Dana-Farber Cancer Institute, initially focused on pancreatic cancer; access runs through Quine Fellows and later Microsoft Discovery.
  • Related models: Co-Scientist, Biomni, STATE

Reading notes

  1. There are no numbers, so its margin over other systems cannot be judged.
  2. The PDAC case shows prediction-to-wet-lab confirmation is feasible, but with no hit rate, negatives or controls it says nothing about other diseases.
  3. The weekend covers only narrowing thousands of compounds to a few; the experimental system rests on years of collaboration.
  4. Restricted access makes independent evaluation nearly impossible.

Caveats

  1. The wet-lab work has no counts, is treated as unverified and cannot be turned into a success rate.
  2. Training data are entirely undisclosed.
  3. Positioned as experimental research technology; the developers warn outputs may be incomplete or inaccurate.

Sources

  1. Microsoft Research official blog (company blog)
  2. Project Ex Vivo page (official site)
  3. About Microsoft Research (official site)

Corrections

None recorded.

Part of the AI for Bio benchmarks survey · all 112 models · open in the explorer. Record as of 2026-11-17.