See the wholedisease in oneafternoon.
DNV-114 is a molecule we designed to stick to cancer cells and light them up on an ordinary hospital scanner. One session maps the disease across the whole body, and it agrees with a surgical biopsy 96.3% of the time. No needle. No two-week wait for the result.
- Agrees with biopsy
- 96.3%
- Injection to answer
- 90 min
- Hospital sites
- 41
The whole picture,
without the scalpel.
A biopsy takes one small sample and assumes the rest of the tumour looks like it. Often it doesn't. DNV-114 attaches to a protein that sits on the surface of these cells wherever they are, so a 90-minute scan shows every deposit at once, including the ones a needle would simply have missed.
Because the result is a number rather than an impression, the same patient can be scanned again a few months into treatment and measured against their own starting point. You can see whether the therapy is working instead of waiting to find out.
- STEP 01
Dose on site
One injection from a ready-made vial, given the way any other contrast agent is. Nothing has to be mixed or manufactured on site.
- STEP 02
Scan at 60 minutes
The patient waits an hour, then spends twenty-two minutes on the scanner the hospital already owns. No new equipment.
- STEP 03
Report the same day
A radiologist gets a whole-body map with a number for each region, in the same system they read everything else in. The patient hears back that day.
Nobody was going to
find this molecule.
For most of the last century, finding a new medicine meant testing millions of existing molecules and hoping one of them happened to fit. It's an expensive way to run a lottery, and for a lot of targets there is simply no winning ticket in the pile. Denova Bio started in 2019 to do the opposite: describe the molecule you need, then build it.
The interesting part isn't the software. It's the loop. Our AI designs candidates, our own lab makes and tests several hundred of them a week, and everything it learns, including everything that failed, goes back into the software by Friday. The design gets better every week rather than every few years.
- Founded, Cambridge
- 2019
- Designs tested weekly
- 600/wk
- Molecules screened at random
- Zero
Plate 01 / The lab, Cambridge
Several hundred designs a week are grown, purified and tested on this bench. The ones that fail are the ones the model learns most from.
How a cell goes wrong
Every cell divides.
Cancer is not an exotic molecule. It is an ordinary process with the brakes off, which is why you find it by reading a surface instead of waiting for a symptom.
Some divide wrong.
A misregulated cell carries the evidence on its membrane long before it is big enough to feel, or to hit with a needle.
The marker is the target.
That marker on the surface is what we designed DNV-114 to grab hold of. We described the grip we needed before any such molecule existed.
So we designed the binder.
It wasn't found in a library. It was designed, built, tested, then tagged so it shows up on a scanner your hospital already owns.
We did not screen for DNV-114. We specified it.
Four steps, repeated week after week. We describe the target, the software designs molecules for it, robots in our lab build and test them, and the results teach the software what to try next. DNV-114 came out of the fourth turn of that loop.
Fig. 03 — binding contacts in the interaction lattice. Points that come within 9 Å of one another resolve as a contact.
Four steps, and the fourth one feeds the first.
- 01Specify
- 02Generate
- 03Build and test
- 04Retrain
06 · Capabilities
Binder design to brief
We make small, antibody-like molecules that grip a chosen target, including targets that decades of conventional methods never cracked.
Radiolabelling chemistry
To show up on a scan, a molecule has to carry a tiny radioactive tag. We plan where that tag goes from the start, so attaching it doesn't ruin the molecule's grip or how fast it leaves the body.
Structure prediction at scale
Before anything is made, separate models predict the shape it will fold into, and we publish how confident they are. Guesses are labelled as guesses.
Automated wet lab
Four robotic lines run through the night without anyone standing over them, which is what makes several hundred tests a week possible at all.
Clinical operations
Getting a hospital up and running is our job, not theirs: training, protocol, paperwork and software. Eleven days on average from signature to first patient scanned.
Data & IP handoff
Partners get the raw measurements and the exact settings we used, so they can check our work rather than take our word for it.
Nothing reaches a hospital that has not been measured first.
A design that looks right on screen is a hypothesis. It becomes a candidate only after a person and a machine agree on what it actually did.
07 · Impact
- Hospital sites
- 41
- Samples tested and sequenced
- 2.10M
- Active programmes
- 14
- Agreement with surgical biopsy
- 96.3%
- Days from contract to first scan
- 11
Figures audited to 31 March 2026. Agreement was measured by scanning and biopsying the same 612 patients and comparing the two results.
“It changed the order of the clinic. We stage on Tuesday and start therapy on Thursday, on the same admission.”
Thirty minutes,
nothing hidden.
One call, whether you run a scanning department, a research programme, or you're simply trying to work out if this is real. A scientist is on every one, and we'll show you what didn't work as readily as what did.
Or write to clinical@denova.bio · Cambridge, UK & Basel, CH