Screening programs
First release
Where carrier screening would help most.
Projects
What we’re building.
One project is live. Many more can start with you.
Live, in development
Scattered genetic data, placed and modeled on one open map.
Next project
More to come
Every project starts as a community idea. Share an idea
Project 1
genomeOS Atlas
Where a genetic variant is common shapes who is offered screening, how a test result is read and where a study can recruit.
The Atlas maps variants on an interactive globe. Pick one, inspect any place and see where evidence is thin.
Open the genomeOS Atlas
- 4,392Variants & alleles in the source tables
- 6.69MPeople represented, not deduplicated
- 2Diseases represented
How we count
Variant total: 4,391 identifiers from the Allele Frequency Net Database (AFND) after the current quality filters, plus the hemoglobin S variant associated with sickle-cell disease (also known as HbS or rs334). Participant total: each retained database population sample counted once, plus people tested in the retained hemoglobin S surveys and male participants in the retained glucose-6-phosphate dehydrogenase (G6PD) deficiency surveys. Some studies may include overlapping groups, so this total describes the scale of the collected evidence rather than a verified census of unique individuals.
Recomputed 2026-09-06 from source snapshot fc17bc1c1d96a0d0766746dcf26277ccdc669717, last updated 2026-08-27.
Why hemoglobin S makes it tangible
One copy can protect against severe malaria; two cause sickle-cell disease. Its frequency traces biology, environment and history.
sickle-cell trait, associated with protection from severe malaria
Arrow keys move a reticle over the map, and the readout below describes the area under it. Home returns to the centre; Escape clears it.
241 confidently above, 622 possibly above
Select an area to inspect.
Simplified from the Atlas’s HbS data.
See it live in the AtlasExpected share of births, from the modeled frequency
- Confidently above: whole range above
- Possibly above: estimate only
- AA: 68–100% of birthsAS: 0–29% of birthsSS: 0–3% of birthsFades away from surveys
- Below 10%
- No estimate
- Survey, sized by people tested
HbS (rs334) · model v3 · data map-2026-08 · revision 6e77ef2 · coarsened for display
View as table
| Share of births | Confidently above | Possibly above | Below, muted |
|---|---|---|---|
| 85% or more | 2,959 | 765 | 496 |
| 90% or more | 2,263 | 1,063 | 894 |
| 95% or more | 1,477 | 852 | 1,891 |
| Share of births | Confidently above | Possibly above | Below, muted |
|---|---|---|---|
| 5% or more | 774 | 1,106 | 2,340 |
| 10% or more | 241 | 622 | 3,357 |
| 15% or more | 18 | 435 | 3,767 |
| Share of births | Confidently above | Possibly above | Below, muted |
|---|---|---|---|
| 0.1% or more | 611 | 921 | 2,688 |
| 0.5% or more | 68 | 520 | 3,632 |
| 1% or more | 0 | 277 | 3,943 |
What it could be used for
Diagnostic context
Future exploration
Regional context before a genetic test.
Trial planning
Where a study could find enough carriers.
The map of what we don’t know
Where the next sample matters most.
Medicines and drug response
Future exploration
Variants that change a dose.
Research and discovery
Patterns worth testing.
How it works
From scattered tables to placed evidence
| Source record | Place ± radius | allele count | alleles tested | Allele orientation | Recruitment | Reuse terms | Outcome |
|---|---|---|---|---|---|---|---|
| City sample | present | present | present | present | present | present | Placed. Every field present; a small stated radius. |
| National survey | present | present | present | present | present | present | Placed, country-wide. Every field present; a radius as wide as the country. |
| Continental group | not a place | present | present | present | present | present | Context only, not a place. Its location is an ancestry label, not a place. |
| Paper table | present | missing | missing: refused | not reached | not reached | not reached | Refused: no count. A percentage with no allele count and no alleles tested. |
| Unplaced label | missing: the build fails | not reached | not reached | not reached | not reached | not reached | Build fails: no place. A population label with no coordinate. |
Every observation needs a place, a radius and a count.
Counts, not guesses
Zero is weak evidence, not absence. Each person carries two copies of the gene; the Atlas counts copies.
Survey 1: 58 of 206 people tested carry the variant: 55 with one copy and 3 with two, 61 of their 412 gene copies. Plausible share of carriers 22.3 to 34.6 percent.
Survey 2: none of 10 people tested carry the variant. Plausible share of carriers 0 to 21.7 percent.
- Carrier
- Not a carrier
- Not tested
Illustrative surveys; 95% Jeffreys intervals.
View as table
| Survey | People tested | Carriers | Gene copies with the variant | Plausible share of carriers (95%) |
|---|---|---|---|---|
| Survey 1 | 206 | 58 (55 with one copy, 3 with two) | 61 of 412 | 22.3–34.6% |
| Survey 2 | 10 | 0 | 0 of 20 | 0–21.7% |
| Survey 2 | 20 | 0 | 0 of 40 | 0–11.7% |
| Survey 2 | 50 | 0 | 0 of 100 | 0–4.9% |
| Survey 2 | 100 | 0 | 0 of 200 | 0–2.5% |
| Survey 2 | 200 | 0 | 0 of 400 | 0–1.2% |
| Survey 2 | 500 | 0 | 0 of 1,000 | 0–0.50% |
From evidence to an explorable map
How the pipeline works
Published samples become counted observations. Counts go straight to the Atlas as measured points; separately, an offline model turns them into a modeled surface with uncertainty. The two arrive as separate layers.
A population-location registry places every observation. Counts go straight to the data service and the map as measured points; separately, frequency and disease-burden models turn them into modeled estimates that pass a validation gate before the data service. The two arrive at the map as separate layers.
- Published samples: Allele counts from papers, theses and databases, in any language.
- Counted observations (AC of AN): Each survey keeps its place, a location radius and its count: AC of AN, never just a percentage.
- Offline model (versioned): Bayesian spatial models are fitted ahead of time and published as versioned results.
- Modeled surface (with 95% range): Every estimate carries a 95% range. Too little evidence means no number, only a hatch.
- The Atlas: Measured and modeled arrive as separate layers you can switch on and off.
- Population-location registry: Where each population was sampled, with a stated radius and provenance.
- Observations (AC of AN): Each survey keeps its place, a location radius and its count: AC of AN, never just a percentage.
- Frequency modeling (Machine learning): Bayesian spatial models, a kind of machine learning, are fitted ahead of time and published as versioned results.
- Disease-burden modeling: Frequency, inheritance, penetrance and population become expected cases, with a range.
- Validation gate (HbS, G6PD, screening): Results are checked against published HbS, G6PD and carrier-screening estimates.
- Data service (no model runs here): Serves versioned results fast. No model runs on this path.
- Interactive public map: Measured and modeled arrive as separate layers you can switch on and off.
Select a part to read its contract.
Six rules the interface cannot hide
Measurements and estimates never merge
Separate tables, layers and legends.
Unknown stays visible
Too little evidence is hatched, never colored.
Zoom cannot manufacture detail
A coarse cell stays one cell.
Assumptions remain explicit
No convenient defaults for places, recruitment or denominators.
Clinical scope is enforced
Single-variant conditions and drug response only.
No evidence means no number
A blank beats a plausible guess.
Where it stands
Available now
- The Atlas: observations and surfaces for 30 variants & alleles
- Population registry and data tools
In active development
- Agentic literature search
- Validation against published estimates
- More datasets
- Disease-burden layers
- Region tools
Future exploration
- An agentic guide citing versioned evidence
- Opt-in personal genomes, behind privacy, consent and governance
- An ancient-DNA time slider
- A cellular atlas