The first open-source coeliac disease digital twin. Glüten screens for undiagnosed coeliac disease from non-specific symptoms, then runs a six-dimension model of the disease to project each patient's trajectory and show exactly where the evidence runs out.
Coeliac disease is not rare. It affects one in every hundred people. It is, however, one of the most underdiagnosed conditions in the world, and the evidence base behind its diagnosis is narrower than most patients realise.
tTG-IgA serology carries higher false-negative rates in Black patients. The best biopsy AI model, at 96% accuracy, was trained exclusively on four NHS trusts in the UK.
Of roughly 32,700 coeliac disease papers in PubMed, about 2,618 study European populations. Only 126 study African populations. Zero validate coeliac diagnostic AI on non-European cohorts.
Symptoms are logged in free-text on Reddit, in WhatsApp groups, in notebooks. 79% of patients use digital media to manage their disease. None of it is structured. None of it feeds back into the science.
Most of the 70% undiagnosed never reach the test. They present with chronic anaemia, unexplained fatigue, early-onset osteoporosis, an IBS label. Individually routine; together a recognisable cluster. Average diagnostic delay: 6–10 years of preventable damage.
One web app. One user: the gastroenterologist, the GP, the dietitian. Two entry points into the same engine — screen a patient with non-specific symptoms, or go straight to the twin with confirmed data. Voice or text in, FHIR-compatible profile out, personalised projection back.
The GP enters non-specific symptoms — chronic iron deficiency, unexplained fatigue, an IBS label, early-onset osteoporosis, family history. Glüten cross-references against ACG, BSG and ESsCD red-flag clusters and returns a coeliac probability plus a recommended next test.
Demographic-aware test advisory. tTG-IgA carries higher false-negative rates in Black patients. If serology is negative but clinical suspicion remains, Glüten flags EMA testing or direct biopsy referral — cited to PMC11308727, not generated.
The clinician speaks or types whatever data they have, serology, HLA type, Marsh score, demographics. Gemma 4 E4B structures it into a FHIR-compatible profile card. 140+ languages. Lagos to Dublin.
Only the layers with data activate. Gemma 4 31B cross-references the patient profile against six dimensions of the coeliac disease model, with PubMed RAG for literature context.
The twin returns a personalised trajectory (projected Marsh, IEL, tTG) alongside per-layer confidence scores that reveal exactly which evidence is thin for this specific patient.
With patient consent, the de-identified structured profile is added back to the disease model under the clinician's institutional governance. Coverage grows. Confidence improves.
Patient profile: African, female, 21, coeliac disease
Glüten is a disease digital twin, not a patient digital twin. It is a composite model of coeliac disease built across six dimensions, drawing on thousands of patients. Each layer is a different view of the same disease, not a separate record for the same person.
When the clinician inputs a patient's data, even partial data from just one or two layers, Glüten runs that profile against the composite model to generate a personalised projection of the patient's trajectory. The per-layer confidence score then tells the clinician exactly how much of the model was relevant to someone like this patient.
A composite of coeliac disease built from six data layers, drawing on thousands of patients across public datasets.
The clinician enters whatever data they have, serology, HLA type, a single biopsy. The twin accepts partial input.
The patient's profile runs against the model. The clinician gets a trajectory prediction plus how confident the model is for someone like this patient.
For a patient with HLA-DQ2.5, Marsh 3b histology, tTG-IgA 84 U/mL, the disease model projects:
marsh 3b → 1 · IEL 42 → <25 per 100 enterocytes · tTG → <20
Glüten is built on curated, citable, open data. No layer is a black box.
Glüten is built by Faith Ogundimu, a first-year PhD researcher at RCSI's Genomic Oncology Research Group, and a coeliac patient. African. Living in Ireland.
When she searched PubMed for studies on coeliac disease in people like her, she found almost nothing. Glüten is a direct response to that absence: a tool designed to measure the gap, and to close it one structured profile at a time.
If the research you need doesn't exist yet, you can help build it.
Glüten doesn't pretend to have every answer. It makes the cost of not having them visible, and gives every consultation a way to close the gap.
