Loadingfonts0%

NeuroNanoTwin
Read the result
Read the result

Hear when
the record moves.

Where it grows first

A $54 BILLION SENSOR MARKET BUILT ON MODELS

Biosensors were worth $31.8 billion in 2024 and should reach $54.37 billion by 2030. Digital twins in healthcare are growing about 26% a year. Every one of those sensors starts life as a computer model, and nobody checks where the model stops matching the real device. ntwin checks.Third-party estimatesMarketsandMarkets, biosensors, 2025 ↗Grand View Research, healthcare digital twins, 2025 ↗

00 / The market

Industry worth

Sensors, the tests they feed, and the computer models used to design them are all growing markets. Nobody yet checks a sensor's model against its evidence before something is built on it. That is the gap ntwin fills, inside these markets.

Third-party estimates, checked 5 September 2026. Endpoints are the sources' own figures; the line between them follows each source's growth rate. Not results of this study.

$100M$1B$10B$100B
May 2025Fujirebio's Lumipulse, the first FDA-cleared Alzheimer's blood test
Aug 2026Roche's Elecsys pTau217, the first stand-alone plasma p-tau217 test cleared
Biosensors$31.8B
Alzheimer's diagnostics$9.94B
Digital twins in healthcare$902.6M
Blood-based Alzheimer's biomarker tests$169.27M

Hover a line to read its source. The log axis keeps a $170 million segment and a $54 billion market on one drawing; the slope is the growth rate.

Why now

  1. 2025 to 2026

    Four FDA-cleared Alzheimer's blood tests in fifteen months

    Fujirebio's Lumipulse in May 2025, then C2N's test on 21 August 2026 and Roche's Elecsys pTau217 on 24 August 2026, the first stand-alone plasma p-tau217 test cleared. p-tau217 is the biomarker this study's sensor targets.

    ALZFORUM, August 2026 ↗

  2. FDA guidance

    Model credibility is becoming a formal requirement

    The FDA's guidance on computational modeling in device submissions is aligned with ASME V&V 40, and in-silico evidence is entering approval dossiers. Someone has to show where a model stops being evidence.

    Exponent on the FDA guidance ↗

  3. 2025

    Digital twins need verification before trust

    A survey in npj Digital Medicine names verification, validation, and uncertainty quantification as the gate for trusting healthcare digital twins, with continuous calibration against real measurements.

    npj Digital Medicine, 2025 ↗

01 / The business

What it could earn

The first thing we sell is the check itself. We take a lab's computer model of a sensor and test where it stops matching the real device, priced per project the way expert reviews are today. Later, labs run the same check themselves with our software. Move the numbers and watch what they build.

Our assumptions, not a forecast. Nobody has paid us yet. One glass block is $10,000.

$18,000
Model checks per year4 · 12 · 30
2027
2028
2029
$6,000
Labs subscribed0 · 5 · 20
2027
2028
2029
2027
$72,000
4 checks · 0 labs
2028
$246,000
12 checks · 5 labs
2029
$660,000
30 checks · 20 labs

02 / The field

Who else is here

The companies that build models sell the model. The experts who check a model do it by hand, once, for a regulator. In Kazakhstan the medical-tech scene is imaging AI, wearables, and antibody design. Nobody here checks a nano-sensor's model against the real experiment before someone builds on it. That is the inner orbit, and ntwin is the only one on it.

Drag to turn the system. Click a planet to fly to it; click empty space to come back.

Kazakhstan · Checks the model

ntwin

Checks a nano-sensor's computer model against the real experiment, automatically, before anyone builds on it.

GlobalKazakhstanntwin

Submitted to AI & Digital Bridge 2026. Astana, 1 to 3 October 2026. Astana Hub Battle finals on 2 and 3 October, a $100,000 prize pool.Astana Hub ↗

NeuroNanoTwin (ntwin) · independent computational study

What happens when AI gets better at the simulation than the experiment?

NeuroNanoTwin, ntwin for short, rebuilt a published p-tau217 nanosensor as a physics model, let optimizers search for a better configuration, then tried to break the answer. The evidence-supported result: no better sensor emerged.

Read the resultHow it was built
Scroll to discover

Fig. 00 · sample drop on the CNT-FET chip · qualitative render

01 / Chapter one

The result

I started this project hoping to design a better p-tau217 sensor before anyone had to build one. I rebuilt a published carbon-nanotube transistor from its physics, then let optimizers loose on it. They found stronger signals almost immediately.

Then I checked where those signals lived. Most of them sat outside the range the real experiment had ever measured. Once I forced the search to stay inside the evidence, the improvement disappeared.

So the honest answer is a negative one. It is not proof that a better sensor cannot exist. It is a clear picture of how an optimizer gets better at the simulation than at the device, and I kept that result instead of hiding it.

Frozen conclusion · Stage 5 report, verbatim

NeuroNanoTwin began as an attempt to computationally improve a p-tau217 CNT-FET nanosensor. The study instead found that unconstrained optimization could exploit experimentally unsupported regions of the computational model. When support constraints and falsification audits were applied, no materially superior robust configuration was identified.

  • In silico
  • Conditional
  • Negative result
  • Frozen record · 88 tests
Stage 4 support landscape · height is P_OOD · derived
hover to read P_OOD · click to route a point to support · 188 of 160,801 supportedyellow valley = supported · red peak = unconstrained optimum

What experimental support means

The Wang experiment only ever measured responses between 0 and 0.3281. That measured range is the envelope.

For every candidate the model predicts a response in 72 assay conditions: nine concentrations times eight ionic strengths. A prediction outside the envelope describes something the experiment never observed, so the model is extrapolating there. P_OOD is the share of the 72 that land outside. A configuration counts as supported when at most 14 of 72 do, which is P_OOD ≤ 20%. No configuration in the whole domain gets below 13 of 72.

…

Stage 4 grid · every dot is one modeled configuration · yellow is supported

160,801conditional configurations modeleddocs/STAGE4_REPORT.md401 × 401 conditional grid · Stage 4 dense reference
188 · 0.1169%met the strict experimental-support rule, drawn here by those 188 cellssupport_threshold_feasibility.csvat P_OOD ≤ 0.20 · none at 0, 5, or 10% · 13 of 72 is the floor
.%

of the unconstrained optimum's fresh conditions fell outside the envelopetable_7_constraint_ablation.csvStage 5 calibration median over 5,000 draws · 181 of 288, 62.85%, at nominal calibration · scale 1.5, 5.0 nm · map derived from the frozen fit

,,
0.00 s
67,359,772 validation evaluations in 6.94 seconds.
  1. 9,180,000B0, B1, B2 under calibration uncertainty on fresh conditions
  2. 26,049,762dense support and objective audit
  3. 32,130,000constraint-ablation calibration audit
  4. 10final fitted-model predictions at the Wang points

validation evaluations in the recorded 6.94 s, zero material improvementsstage5_summary.jsonmeaningful model-condition evaluations · 6.94 s wall time in the recorded environment · about 9.7 million per second, DERIVED

Read the full record

02 / How it was built

Inside the model

Scroll to open the device. Every layer is one step of the chain from concentration to current, and every number shows where it came from.

  1. 01 / 07

    Sample drop

    A drop of buffer carrying synthetic p-tau217. The model treats it as 0.01x PBS at 1.63 mM ionic strength, a declared scenario rather than a measured value, which sets a Debye screening length of about 7.5 nm.

    • ASSUMED_SCENARIO
    • 0.01x PBS
    • 1.63 mM
  2. 02 / 07

    Recognition

    Antibodies on gold nanoparticles capture the peptide. Occupancy follows a Langmuir curve with an apparent K_D of 47.5 fM, fitted to five digitized Wang points, not measured on the device.

    • FITTED
    • K_D,app 47.5 fM
    • 5 digitized means
  3. 03 / 07

    Bound charge

    Each captured peptide adds charge to the surface, taken as +4e per molecule, at an effective distance of 10 nm from the channel. Both are transferred or assumed, never measured here.

    • ASSUMED_SCENARIO
    • +4e
    • 10 nm, range 5 to 13
  4. 04 / 07

    Screening

    Ions in the drop screen that charge over the Debye length. Higher ionic strength means shorter reach and a weaker signal. In the model, ionic strength explains 80.5% of the response variance.

    • DERIVED
    • Sobol total 0.805
  5. 05 / 07

    CNT channel

    Whatever potential survives reaches the nanotube network and shifts its conductance. The response is 17.24 per volt times the effective gate shift, the model's second and last fitted number.

    • FITTED
    • 17.24 V⁻¹
    • local linear transduction
  6. 06 / 07

    Read-out

    Source and drain read the normalized current change. Five Wang points, 0.0572 to 0.3281, anchor the whole chain. Contact resistance, transconductance, and channel geometry stay unknown.

    • DIGITIZED_FROM_FIGURE
    • UNKNOWN
    • R² 0.931
  7. 07 / 07

    The device

    A drop of sample sits on the chip and the transistor's current is read electrically, with no labels and no optics. That is what makes a bench assay a candidate for a portable test, and why an honest model of it matters. The public record gives a gate bias of −0.5 V, a calibrated range of 3 fM to 30 pM, a published detection limit of 1.66 fM this project did not re-verify, 4.8% spread across nine sensors at 30 fM, and about a 10% current drop after seven days at 4 °C. Channel geometry and chip dimensions are not in that record.

    • PUBLISHED
    • LOD 1.66 fM not re-verified
    • geometry UNKNOWN

03 / Why it matters

Nanosensors could turn a lab assay into a finger-prick test.

04 / What this study adds

Carbon-nanotube transistors read biomarkers electrically, with no labels and no optics. Wang et al. reported a 1.66 fM detection limit for p-tau217, a published figure this project did not re-verify.

Knowing when a model of that sensor can be trusted is the hard part. NeuroNanoTwin tested it: of 160,801 modeled configurations, 188 met the strict experimental-support rule, and none was a materially better sensor.

  • PUBLISHED LOD 1.66 fM · not re-verified
  • concept object · not a product
01 / Open

A drop in, a number out.

Sample enters at the well, a channel carries it to a chamber over the chip, and the chip reads the biomarker as a change in current. Only the chip is a published device. The rest of this object is a concept drawn for this page.

  • CONCEPT OBJECT
  • NOT A PRODUCT
02 / The chip

The real part.

Carbon nanotubes bridge two gold electrodes. Gold nanoparticles on the tubes carry antibodies that catch p-tau217, and each capture shifts the current. Wang et al. built and measured this architecture. Its channel geometry and chip dimensions were never published, so nothing here is to scale.

  • PUBLISHED ARCHITECTURE
  • GEOMETRY UNKNOWN
  • NOT TO SCALE
03 / The parts

Seven parts. One of them published.

Drag to turn it. Click a part to jump to it. Every part except the chip is a concept.

03 / Kazakhstan

Where it grows first

The whole thing in one paragraph

ntwin began as a computational study of one device: a carbon-nanotube transistor meant to detect p-tau217, the blood marker now used to diagnose Alzheimer's disease. We built a physics model of it from five digitized points in a published experiment, then asked an optimizer for a better sensor and spent 67,359,772 model-condition evaluations on the answer. The optimizer found stronger simulated signals and the evidence refused to back them: of 160,801 modeled configurations only 188 met the strict experimental-support rule, and at fresh conditions the unconstrained optimum fell outside the measured envelope in 62.15% of cases. We then tried five separate ways to break that verdict and failed every time, so the result is a strong negative one and it is frozen exactly as it came out. What is worth building a company on is the machinery rather than the device: a way to take any sensor's computer model and show where it stops matching the real experiment, before a laboratory builds on it, a hospital buys it, or a regulator reads it. Kazakhstan is where that starts, because the next step needs things that are already here. The study was written in Almaty and submitted in Astana. The laboratory that grows carbon nanotubes is at al-Farabi KazNU, the university hospitals and the Alzheimer's cohort that could supply real plasma are in Astana, Almaty and Karaganda, the national registry recorded 496 Alzheimer's cases in 2021 and 3,706 in 2025, and device registration is moving to one Eurasian procedure that will have to judge evidence made of simulation. So the plan is three steps: put the model on a Kazakh bench, run it in a Kazakh hospital, then check other people's models before they reach a ward.

A plan, not a contract. Every institution named is a candidate; none has been contacted or has agreed. Third-party facts checked 5 September 2026; none is a result of this study.

  1. 01 / 05 · The ground

    The demand is already in the registry

    Kazakhstan's health registry counted 496 Alzheimer's cases in 2021 and 3,706 in 2025, inside 11,087 dementia cases. The Ministry's own experts read the rise as ageing plus better recognition and registration, which is the honest reading; either way an earlier, cheaper test is wanted here before it is wanted anywhere else.

    • Salidat Kairbekova National Scientific Center for Health Development
    • Reported by The Astana Times, March 2026

  2. 02 / 05 · 01 · 2026 to 2027

    Put the model on a Kazakh bench

    The frozen record ends with one recommended experiment: three antibody-loading protocols, three ionic strengths, five concentrations, on devices held out from calibration. Step one is a lab that can build or source CNT-FET devices and run that plan with us, so the model is checked against measurements nobody fitted it to.

    • National Nanotechnology Laboratory of Open Type, al-Farabi KazNU (carbon nanotube synthesis, services to third parties)
    • National Laboratory Astana, Nazarbayev University
    • NU Laboratory of Biosensors and Bioinstruments

  3. 03 / 05 · 02 · 2027 to 2028

    Run it with real plasma, inside a hospital

    Once the bench experiment says where the model holds, the same plan runs on patient plasma. That needs a memory clinic, an ethics board, a biobank, and the clinicians who already recruited a Kazakh cohort. The hospital gets a sensor checked against its own samples before anyone calls it a test.

    • University Medical Center, Astana (778 beds, JCI academic medical centre since 2024, University of Pittsburgh partner)
    • Astana Medical University, Department of Neurology
    • Medical University of Karaganda, Institute for Life Sciences
    • National Center for Biotechnology, Astana (gene diagnostics, 13 laboratories)

  4. 04 / 05 · 03 · 2028 onward

    Check models before they reach a ward

    By then the check is a product: any sensor or twin that comes with a computer model gets checked against its own evidence before a hospital buys it, a regulator files it, or a student trusts it. Kazakhstan's medical-AI infrastructure and the Eurasian registration route are where that service lands first.

    • HAQ MedHub, Astana Hub's medical-AI infrastructure (launched July 2026)
    • National Center for Expertise of Medicines and Medical Devices
    • Asfendiyarov Kazakh National Medical University, Almaty
    • Kazakhstan Alzheimer and Dementia Alliance

  5. 05 / 05 · The first stop

    Astana, 1 to 3 October 2026

    This site is the submission to AI & Digital Bridge 2026, where the Astana Hub Battle finals run on 2 and 3 October for a $100,000 pool. Residency, the young-scientist grants, and the commercialisation competition are the three routes that could pay for the bench and the ward.

    • Zhas Galym young-scientist grants
    • Science Fund commercialisation grants
    • Astana Hub residency

Who is where

  • Astana51.17° N, 71.43° E

    • Nazarbayev University, National Laboratory Astana
    • University Medical Center
    • Astana Medical University
    • National Center for Biotechnology
    • National Centre for Neurosurgery
    • Astana Hub, HAQ MedHub

  • Almaty43.24° N, 76.95° E

    • al-Farabi KazNU, National Nanotechnology Laboratory of Open Type
    • Asfendiyarov Kazakh National Medical University
    • Haileybury Almaty, where the study was made

  • Karaganda49.80° N, 73.10° E

    • Medical University of Karaganda, Institute for Life Sciences

  • Shymkent42.32° N, 69.59° E

    • UMC Heart Center Shymkent (since 2024)
    • Shymkent Hub, an Astana Hub branch

In a Kazakh hospital

  1. 01

    Before a hospital buys a new sensor test

    A laboratory or a procurement office is offered a biosensor test that comes with a computer model of the device. ntwin checks where that model stops matching the vendor's own evidence, and hands the buyer a report they can read.

  2. 02

    Before a plasma sample is spent

    Patient plasma is scarce. The check ranks the conditions where the sensor is least certain, so a memory clinic spends its samples where they teach the most: the recommended three-by-three-by-five plan instead of a guess.

  3. 03

    Before a device is registered

    A dossier for the Kazakh or Eurasian route that leans on simulation has to show where the model holds. The check writes that section, with the envelope, the held-out conditions, and the exact figures the reviewer asks for.

  4. 04

    In a lecture hall

    Medical and engineering students rerun the whole check from the frozen record in one command; the recorded run takes 6.94 seconds. A negative result that survived five attempts to break it teaches more than a promise.

What could pay for the bench and the ward

  • Zhas Galym young-scientist grants

    The Ministry of Science and Higher Education is funding 1,000 research grants and 250 international internships for young scientists in the 2026 to 2028 cycle.

    The Astana Times, August 2026 ↗

  • Science Fund commercialisation grants

    The Science Fund runs the Committee of Science's competition for commercialising research results, and now also accelerates, incubates, and venture-finances scientific projects.

    JSC Science Fund, news ↗

  • Astana Hub residency

    Residents are exempt from corporate income tax, VAT, individual income tax, and social tax anywhere in Kazakhstan, and get accelerators, grants, and the partner network. The study is submitted to Astana Hub's Digital Bridge 2026.

    Astana Hub for startups ↗

First stop: AI & Digital Bridge 2026. Astana, 1 to 3 October 2026. Astana Hub Battle finals on 2 and 3 October, a $100,000 prize pool.

Institution sourcesal-Farabi KazNU, National Nanotechnology Open Laboratory ↗University Medical Center, about ↗Nazarbayev University School of Medicine ↗National Center for Biotechnology, about ↗National Centre for Neurosurgery, about ↗Kazakhstan Alzheimer and Dementia Alliance (ADI member) ↗The Astana Times on HAQ MedHub, July 2026 ↗