Finding cancer early changes the odds.
Pancreatic cancer, five years on: 44 in 100 alive if found early, 3 if found late.
We believe nobody should learn they have cancer when it is too late. Today's tests can't see it that early: the sample holds too little of its DNA.
Scroll ↓
Tumours shed DNA into the blood
Dying tumour cells release short fragments of DNA. Catch them and you can read the tumour without cutting into it: a liquid biopsy.
The fragments thin out in five litres of blood. At one in a million, the tube expects 0.006 tumour copies of any one mutation. Usually zero.
Zviran et al., Nature Medicine 2020: median 6,000 copies in 194 patients. “If no physical fragment that supports a targeted [mutation] is present in the sample, even ideal ultra-deep targeted sequencing will fail to discover a signal.”
What if we just had more material?
More molecules.
Our device, HFX, pulls the DNA from a whole morning void, collected at home: 50× what a blood draw holds. A DNA test downstream gets more to read.
Genome copies per sample. We measure 250–1,000 ng of DNA per 100 mL of urine; a morning void is ~300 mL.
Urine carries DNA from the blood too
The kidneys pass DNA fragments from the blood into the urine.
You can collect a whole cup at home, every year, with no needle.
Keep the DNA. Leave the water.
A whole void goes in, 100–500 mL. One small pellet of DNA comes out: 50× what a blood draw holds.
The sample is sent from home and keeps for 8 days in our preservative. The pellet is DNA in a few drops, ready to be sequenced.
Measured in healthy donors: 250–1,000 ng DNA per 100 mL, rising in step with volume up to 160 mL. A morning void is ~300 mL (HFX takes up to 500 mL). 3.3 pg per copy; a blood sample holds a median 6,000 copies. Counted from DNA mass; the usable human share is part of what we measure first.
Now there's something to find
At one tumour copy in 100,000, a blood tube expects 0.06 tumour copies. The cup expects 3, if urine carries the same tumour share as blood.
More copies, more chance one is there
Chance that at least one copy of a given tumour mutation is in the sample
At 1 in 100,000: blood 6%, HFX 89% to ~100%. At 1 in a million: 0.6% vs 20–59%.
The open question: urine also carries DNA from the kidneys and bladder, so its tumour share may be lower than blood's. At a tenth of blood's share (dashed), the cup still gives 20–59% against blood's 6%.
For pancreatic cancer, that share is the first thing we measure.
A home test for pancreatic cancer.
Half are found after the cancer has spread. No screening test is recommended for the public. 52,740 Americans die of it every year.
We start with people who are already being watched: families at high risk, new diabetes after 50, and patients after surgery, where the new RAS drugs are heading.
First pancreatic data: Q2 2027.
Five-year survival localised 44%, distant 3% (SEER). US deaths 2026: ACS.
We make the data easy to collect
A blood test needs a clinic, a needle and a nurse. A cup needs a bathroom. So the same person can give a sample every year, or every month in a trial.
More molecules per sample. More samples per person. More people.
The models that read cancer DNA learn from thousands of deep samples. Our studies collect whole voids with matched blood and the hospital's diagnosis.
Each dot: one cup. 100 people, three years.
The tests are ready. The sample isn't.
Reading DNA got a million times cheaper. People take the easy test. Medicare now pays for early detection. Every test is still capped by the blood draw.
Reading got cheap. The sample stayed the same.
Reading a human genome cost about $100 million in 2001 and about $60 today. A blood draw held ~6,000 copies then, and holds ~6,000 now.
Blood tests already choose the drug (Guardant360), watch for relapse (Signatera) and screen for bowel cancer (Shield): ~600,000 tests last quarter from four US companies, up 70% in a year.
Genome cost: NHGRI sequencing-cost data, approximate; $60 genome: Ultima Genomics, October 2026 (company figure). Tests: Q3 2026 vs Q3 2025, company reports: Natera, Guardant (incl. Shield), GRAIL, Tempus.
The record numbers need the tumour first
The best relapse tests now claim to see one tumour fragment in a million. They sequence the patient's tumour first, then track up to 5,000 of its mutations across the copies in the tube. Lots of chances from few copies.
Screening has no tumour to learn from. Screening tests stop at about 1 in 2,000 to 1 in 600.
To find a cancer nobody knows about yet, you need more molecules.
Company claims, 2026: Abbott Oncodetect (under 1 part per million at high input, up to 5,000 variants); Signatera (Natera whitepaper); NeXT Personal (Oncotarget 2024); Shield (FDA SSED P230009); GRAIL (PLoS One 2023).
Cancer drugs now aim at mutations
The FDA approved daraxonrasib, the first RAS drug for pancreatic cancer. RAS mutations drive about 9 in 10 of these cancers.
Doctors read tumour DNA to choose these drugs and to see if they work: in 28 of 28 patients on daraxonrasib it fell by half or more.
The RAS drugs are now in trials right after surgery. That's where the blood holds the least tumour DNA to read.
RASolute 302 (ASCO 2026, NEJM); FDA approval 26 Aug 2026. KRAS: ~93% (TCGA 2017). ASCO Post, May 2026 (28 patients). RASolute 304 (adjuvant), Revolution Medicines Q3 2026.
Make the test easy and people take it
Guardant sold 83,000 Shield blood tests last quarter, 3.5× a year earlier. People bought 185,000 Galleri tests in 2025, mostly without insurance.
A cup at home is easier still: no needle, no appointment, every year.
Randomised trial, Gut 2024 (30.5% vs 13.0%; Guardant-funded). Guardant Q3 2026; GRAIL FY2025.
Medicare now pays to find cancer early
It already pays $1,495 per Shield test and $3,500+ per Signatera test.
Watching people at high risk pays off: about €14,000 per healthy year gained in one high-risk group.
Public Law 119-75 (MCED Act, 3 Feb 2026). CMS ADLT rates. Leiden CDKN2A surveillance, cost per QALY.
Measured on patient urine.
Every bladder-cancer sample gave more DNA than a blood draw, from 85 mL. That shows yield. The pancreatic signal is what we measure next. Patent filed.
DNA recovered per sample. HFX: six urine samples (85 mL) from four bladder-cancer patients, our 2026 poster. Grey: a 300 mL void at the same yield per mL, not yet measured; yield rises in step with volume up to 160 mL. Kit and blood: typical yields.
What we know, and what we don't
- Shown by us. Healthy donors: 250–1,000 ng of DNA per 100 mL of urine, so about 50× a blood draw from a 300 mL void. Bladder-cancer patients: 500–4,080 ng from 85 mL. A first targeted-sequencing run worked.
- Published by others. Pancreatic tumour DNA reaches urine. In 56 patients, KRAS was found as often in 3 mL of urine as in blood: 48% each.
- Not known yet. Whether a whole void finds pancreatic cancer early. There is almost no stage I data. Our study's first data: Q2 2027.
HFX yields, internal data. Terasawa et al., Cancer Biol Ther 2019 (stage II–IV, no healthy controls).
PRECIS-BC: can a urine test spare the bladder?
About 1 in 3 patients with muscle-invasive bladder cancer have no tumour left after chemotherapy. Scans can't tell who they are, so nearly all lose their bladder.
With Radboudumc, we test whether tumour DNA in urine, read through HFX, can find the patients who could keep it. 60 patients, 2026 to 2028, funded by Health Holland.
Why not just…
- Draw more blood? Tests already take up to four tubes. That is about 24,000 copies, against about 300,000 from one cup.
- Use a bigger urine kit? Kits take 10–50 mL. A few products claim more, without convincing published data. HFX takes the whole void.
- Concentrate urine in the lab? It works: agreement with the tumour rose from 73% to 89% at 90–110 mL (19 patients). That was a manual lab protocol. HFX makes it one routine step.
Shield, up to 4 tubes (FDA SSED P230009). Fujii et al., Clin Cancer Res 2017 (mixed advanced cancers).
Team
Advisors
Each has taken a life-science company to a sale.
More molecules. Let's talk.
The whole field makes the reading better. We make the sample bigger.
Next: first pancreatic data, Q2 2027.
Investors: [email protected] · Projects and trials: [email protected]
Hologenomix B.V. · Planet B.io, Biotech Campus Delft