NEO Bio n Techby NEO AI
EN
Log in Request access

In development - a NEO AI vertical

Computational biologythat shows its evidence.

We are building a research environment that turns public biological data into original analyses, testable hypotheses and reproducible results. Every result will leave with its sources, its limits and a way for anyone to check it.

Explore the programmes See how results are checked

Artistic illustration, not a structure. Drag to turn it.

We do not own a laboratory. We own a method: public data, protocols fixed before the test, and results sealed with everything needed to check them. When a hypothesis needs an experiment, it goes to a partner laboratory under a recorded protocol.

Three research programmes

Each programme asks narrow, measurable questions of existing public data. Each states what computation can establish and what would need a laboratory.

Synthetic data, illustrative
  • Sourced link
  • Inferred by a model
  • Contradicting evidence

Diseases, genes, targets, molecules, studies

Discovery Atlas

Maps the evidence around a disease or a target. Every link carries its source, its version and its limits, and contradictions stay visible instead of being averaged away.

Questions it asks

  • Which lines of human evidence support or contradict a target for a defined indication?
  • How much of an apparent consensus comes from independent studies, and how much from the same dataset reused?
  • Which hypotheses remain under-explored after a documented prior-art search?

What it will deliver

Sourced evidence maps, hypothesis cards with their counter-arguments and validation plans, and dossiers sealed with their proofs.

Synthetic data, illustrative
  • Reference outline
  • Model mask
  • Failure case

Biological images, segmentation, morphology

BioVision

Measures how image-analysis models really behave on biological images. Models are compared on exactly the same protocol, with data split by plate, laboratory or donor so that test images never leak into training.

Questions it asks

  • Does a segmentation model keep its accuracy on images from a laboratory it has never seen?
  • Do popular benchmarks overstate performance because training and test images share a source?
  • Which failure modes dominate, and what do they cost?

What it will deliver

Benchmarks with leakage controls, reproducible reports, model cards and atlases of failure cases.

Synthetic data, illustrative
  • Condition A
  • Condition B
  • Low confidence

Single cells, perturbations, generalisation

Cell Intelligence

Studies single-cell and perturbation data, and checks whether a model's predictions hold on perturbations or cell contexts it has never seen, against simple baselines first.

Questions it asks

  • Does a candidate model beat simple baselines on perturbations it has never seen, within uncertainty?
  • Which signatures stay consistent across studies, donors and batches?
  • Where do models fail: by cell type, by perturbation, by dataset?

What it will deliver

Baseline comparisons with confidence intervals, model cards, and negative results published with the same visibility as positive ones.

An illustration is not a structure

The helix at the top of this page is art. The molecules here are real: atomic coordinates measured by X-ray diffraction and published in the Protein Data Bank. We keep the two apart everywhere, and every scientific view carries its provenance.

These structures are public reference data, not results of NEO Bio n Tech. The file served here is byte-identical to the RCSB copy; its fingerprint is recomputed by your browser.

PDB entry
1BNA
Molecule
DNA 5'-CGCGAATTCGCG-3', Drew-Dickerson dodecamer
Method
X-ray diffraction
Resolution
1.9 Å
Deposited
1981-01-26
Primary citation
Drew HR et al., Proc Natl Acad Sci USA 78:2179-2183 (1981)
Licence
CC0 1.0, wwPDB archive
Atoms drawn
-
SHA-256 of the file
df42f1506792f191b957227b061360652adcf6f813eb69d9ec553067ea584670
Colour by element: carbon, nitrogen, oxygen, phosphorus, sulphur. C N O P S

From a question to a result you can check

Every project follows the same path. Each step leaves a dated trace, and the steps that need a human decision cannot be skipped by software.

  1. 1

    Ask a precise question

    A named scientist owns a testable question. Prior work is searched first, and the sources consulted are listed.

  2. 2

    Fix the protocol first

    Comparators, metrics, data splits and stopping rules are versioned before the final test data is touched.

  3. 3

    Compute with provenance

    Every run records its data versions, code, model, parameters, seeds and environment. Failed runs and negative results are kept.

  4. 4

    Review, seal, publish

    A reviewer other than the author evaluates a precise version. The evidence is sealed and time-stamped, and only authorised versions become public.

Integrity is not truth

Each result carries four separate statements. They are never merged into one green badge, because a result can be intact and still unreproduced, or reviewed and still unpublished.

Technical integrity

The files match their recorded fingerprints; the signature and the timestamp check out.

It does not say whether the analysis is right.

Reproducibility

Someone reran the analysis from the sealed dossier and obtained the same output, within a stated tolerance.

Reproducing a flawed design reproduces the flaw.

Scientific validation

The level of evidence: computational only, or confirmed by an experiment, and by whom.

Computation alone never establishes a treatment effect or a mechanism.

Publication status

Internal, preprint, peer reviewed, corrected or withdrawn.

A preprint is not peer reviewed; peer review is not replication.

Change one character, and the seal breaks

A proof dossier is sealed by fingerprinting each of its parts with SHA-256, then combining the fingerprints in a Merkle tree. Only the root of that tree will be time-stamped and anchored on a public blockchain. Edit any part below and watch which fingerprints change.

Computed live in your browser with Web Crypto. Nothing is sent anywhere.

Example dossier for illustration
Merkle root -
-
-
Root at the moment of sealing: -

What a blockchain anchor proves

It proves

That a precise fingerprint existed at a given block time, and so did every dossier included in it.

It does not prove

That a result is true or new, who the real author is, who owns the rights, or that a patent exists.

No content, personal data or secret is ever written on a chain: only a fingerprint and a format version.

LILI works in the open

LILI, the assistant of NEO AI, is designed to help our scientists search the literature, prepare protocols, launch pipelines and assemble dossiers. Every step it takes will be visible: the plan, the sources, the tools called, the estimated cost, and what still needs a human.

An answer from LILI is never a validation, and an agreement between several AI agents never counts as an independent review.

No AI system is an author. Uses of AI are disclosed on every public result.

One brain for every vertical

NEO Bio n Tech is a vertical of NEO AI, the intelligence operating system of My NEO Group. NEO AI has six verticals today, twenty more in the launch pipeline, and about a hundred in the end. Each vertical keeps its own data and permissions.

Discover the NEO AI verticals

LILI

Illustrative sequence

Compare three candidate targets for this indication and show the contradicting evidence.

  1. PlanSearch prior work, collect human genetic and expression evidence, list contradictions, draft a comparison.
  2. SourcesLiterature, genetics and target databases, each with version and licence.
  3. ToolsLiterature search, evidence graph, deduplication of studies.
  4. Estimated costShown before the run, capped by the project budget.
  5. Still neededReview by a scientist other than the author before anything is shared.

Built on public science

Before writing any code, we qualified the public sources the programmes will draw on: their access, their licence for each use, their volumes and their overlaps. A source whose rights are unknown is not used.

  • Literature

    PubMed, PubMed Central, Europe PMC, BioStudies

  • Genetics

    Open Targets, GWAS Catalog, ClinVar, gnomAD, GTEx, Ensembl

  • Molecules

    ChEMBL, PubChem, BindingDB, ChEBI

  • Proteins and pathways

    UniProt, RCSB PDB, AlphaFold DB, InterPro, Reactome, STRING

  • Cells and perturbations

    GEO, CELLxGENE, scPerturb, Tahoe-100M, Open Problems

  • Imaging

    BioImage Archive, IDR, BBBC, Cell Painting Gallery

  • Trials

    ClinicalTrials.gov, openFDA

Preliminary analysis completed in October 2026: 110 sources qualified, 26 datasets sized, 61 models and libraries reviewed for the licence of their code and weights, and 49 live access tests run on public interfaces.

Names shown are public resources in our catalogue. Listing a resource does not imply a partnership or an endorsement.

Where we are

NEO Bio n Tech is in development. Nothing on this page is presented as a discovery.

  1. Done

    Analysis of sources, licences and models

    Completed in October 2026, with live tests of public interfaces.

  2. Done

    Research environment designed

    Design system, 3D visual library and navigable prototype with demonstration data.

Computed here, tested elsewhere

QuestionWhat computation can establishWhat needs an experiment
Is a gene associated with a disease in the sources consulted?Yes, with sources, versions and populations statedNo
Does modulating a target change disease biology?Only a hypothesis, with supporting and contradicting evidenceYes, in a recorded protocol with a partner laboratory
Does a segmentation model hold up on a new laboratory's images?Yes, on held-out public data split by sourceOnly if no independent public data exists
Does a predicted perturbation response hold in real cells?Agreement with held-out public measurementsYes, for any new prediction

Work with us

Universities, laboratories, data providers and research funders: if you want to test a hypothesis, replicate a result or share a dataset under clear terms, write to us through the contact page of My NEO Group.

Contact My NEO Group Request access