How a Personalized Cancer Vaccine Works: from tumor DNA to mRNA.

The short version: a data pipeline that ends in a drug

Most cancer drugs are designed once and given to everybody. A personalized neoantigen vaccine is designed from scratch for a single patient. If you strip away the jargon, it is really a data pipeline that ends in a drug, and it runs in four steps:

1. Tumor DNA. Read the genetic code of the cancer, and of the dog's healthy tissue, and compare them.
2. Somatic mutations. Keep only the changes the tumor has and the healthy cells do not.
3. Predicted immune targets. Work out which of those changes the dog's immune system can actually recognise.
4. An mRNA. Write a set of instructions that teaches the dog's T-cells to hunt down cells carrying those exact changes.

Everything else in this article is detail on those four arrows. The thing worth holding onto is that no step invents anything. The vaccine is not designed to attack the cancer in some clever new way. It simply points at something the cancer was already carrying, and hands that description to the immune system.

Step 1: read the tumor's DNA

Cancer begins when a cell's DNA, its instruction manual, picks up mistakes. Those mistakes are called mutations, and they accumulate as the tumor grows. The tumor's manual is therefore a slightly misspelled copy of the dog's original.

So the process starts with two samples, not one. A piece of the tumor, and a sample of healthy tissue, usually a simple blood draw. A sequencing lab reads the DNA in both. Reading one without the other would be almost useless, and the reason why is the single most important idea in this whole field.

Step 2: find the mutations only the cancer has

Every dog is born with thousands of harmless quirks in their DNA. Compared against some generic reference genome, a healthy dog looks "mutated" in a great many places. Those inherited quirks are present in every cell in their body, including the healthy ones.

Comparing the tumor against the dog's own healthy DNA strips all of that away. What survives the comparison is a much shorter list: the changes that exist in the cancer and nowhere else in the animal. These are called somatic mutations, and they are the tumor's weakness.

So what exactly is a neoantigen?

A mutation in DNA is only interesting if it changes something the immune system can see. Many do not. But some change the recipe for a protein, so the cancer cell builds a protein that is very slightly wrong. A single amino acid swapped for another, like one misspelled word in a long book.

Cells routinely chop up their own proteins and display the fragments on their surface, like holding up samples of everything being made inside. When a cancer cell displays a fragment of one of those misspelled proteins, it is displaying something that does not exist anywhere in the healthy body.

That fragment is a neoantigen. "Neo" for new, "antigen" for something the immune system can recognise. You can think of it as a name-tag that only cancer cells wear, one that says "this cell is abnormal."

The immune system is very good at ignoring the body's own proteins. It has to be, or it would attack healthy tissue. A neoantigen is interesting precisely because it is not one of the body's own proteins. It is foreign, in the same way a virus is foreign, even though it was made by the dog's own cell.

Step 3: work out which targets this dog can actually see

Here is where it becomes genuinely personal, and not just marketing language.

Cells do not wave protein fragments around loose. They hold them up in a molecular display case. In dogs this display system is called DLA (Dog Leukocyte Antigen). It is the canine counterpart of the system people have, which is called HLA.

The catch is that display cases come in different shapes, and every dog inherits their own set. A fragment that slots neatly into one dog's display case, where a patrolling immune cell can inspect it, might not fit into another dog's at all. If it cannot be displayed, the immune system never sees it, and it is useless as a vaccine target no matter how tumor-specific it is.

So the software does two jobs. It reads the list of somatic mutations and predicts which ones produce a neoantigen at all. Then it predicts which of those will actually fit this particular dog's display case. What comes out is a ranked shortlist of targets, chosen for one specific animal.

Why specificity is the whole safety story

It is tempting to treat safety as a separate box to tick after the science is done. In a neoantigen vaccine it is not separate. The specificity is the safety, and the specificity is also the efficacy. They are the same property viewed from two sides.

Think about what you are asking the immune system to do. You are training it to find and destroy cells carrying a particular molecular flag. If that flag also appears, even faintly, on healthy tissue, you have just trained the immune system to attack the dog. That is the entire risk of the approach, stated plainly.

The defence is the same comparison from step 2, taken seriously. A target only qualifies if it comes from a mutation present in the tumor and absent from the dog's own healthy DNA. Anything that is not unique to the cancer gets discarded before it is ever considered as a vaccine target. It is a subtractive process, and most candidates do not survive it.

Which is also why the approach can be so precise when it works. A conventional chemotherapy is a blunt instrument that harms fast-dividing cells wherever it finds them, cancerous or not. A neoantigen vaccine is aimed at a description that, if the pipeline did its job, matches nothing in the healthy animal.

Step 4: write the recipe in mRNA

Now you have a shortlist of flags. You need to show them to the immune system in a way it takes seriously. That is what the mRNA does.

mRNA is the body's own short-term messaging format. It is a temporary instruction note that a cell reads, acts on, and then throws away. Your cells make and destroy mRNA constantly, all day. It does not enter the DNA and it does not stay.

The vaccine is an mRNA that spells out the chosen neoantigens. When it is injected, the dog's own cells read the note, build those exact protein fragments, and display them. The immune system encounters them in a context that says "pay attention," learns the pattern, and goes looking for anything else wearing the same flag. What it finds is the tumor.

It is worth saying clearly: the vaccine does not contain cancer, and it cannot cause cancer. It contains a description of a few protein fragments, written in a language the body erases after reading.

Does this actually work? The human evidence

This is the fair question, and the honest answer is that the human results are genuinely encouraging, still early, and worth reading carefully rather than in a headline.

Melanoma

The most advanced example is mRNA-4157 (also called V940), from Moderna and Merck. In a randomised trial in people who had had high-risk melanoma surgically removed, patients received either the standard immunotherapy drug pembrolizumab (Keytruda) on its own, or pembrolizumab plus a personalized neoantigen vaccine built from their own tumor.

Adding the vaccine lowered the risk of the cancer coming back, or of death, by roughly 44 percent compared with the immunotherapy drug alone (a hazard ratio of 0.561). At 18 months, 79 percent of the vaccinated group had not relapsed, against 62 percent of the group on the drug alone. Longer follow-up presented at a major oncology conference showed the separation holding: recurrence-free survival at two and a half years was about 75 percent versus 56 percent. (Weber et al., The Lancet, 2024)

Two caveats belong right next to that number, and we would rather state them ourselves. This was a phase 2b trial with 157 patients, which is modest. And the result, while clearly favourable, sat just at the edge of conventional statistical significance. A much larger phase 3 trial is now running to settle the question. This is a promising signal, not a closed case.

Pancreatic cancer

The second example is autogene cevumeran, from BioNTech and Genentech, tested after surgery in pancreatic cancer, a disease with famously few mutations and a famously poor outlook. In a phase 1 trial of 16 patients, the vaccine induced new neoantigen-specific T-cells in half of them. (Rojas et al., Nature, 2023)

What happened next is the striking part. At more than three years of follow-up, the patients who mounted a T-cell response had still not reached their median time to recurrence, while those who did not respond relapsed at about 13 months. The vaccine-induced T-cells were still present years later, with an estimated average lifespan of nearly eight years. (Sethna et al., Nature, 2025)

Again, the caveat matters: 16 patients, and the comparison is between people who happened to respond and people who did not, rather than between randomly assigned groups. It shows the immune mechanism working and lasting. It does not, on its own, prove the vaccine causes the survival difference.

What is different about dogs

Less than you might expect. The logic transfers almost intact, and three things change:

• The reference genome. Mutations are called against a canine genome rather than a human one.
• The display system. Dogs use DLA where people use HLA, so the prediction models have to be built for canine biology rather than borrowed from human data.
• The codon table. The mRNA is optimised to be read efficiently by canine cells.

That is genuinely it, in terms of shape. Dogs also happen to be an unusually good place to do this work: they develop cancers spontaneously, live alongside us in the same environment, and their tumors behave much more like human tumors than anything that happens in a laboratory mouse.

What this is not

We would rather be straight with you than sell you something.

This is an emerging, experimental approach. It is not a cure, and nobody can promise you an outcome for your dog. In the United States, veterinary biologics are overseen by the USDA's Center for Veterinary Biologics rather than the FDA, and a personalized vaccine like this is not an approved product you can simply ask for.

It is not a replacement for surgery, radiation, or chemotherapy where those are the right treatment. The human trials that produced the results above were all run in the adjuvant setting, meaning after the main tumor had been removed, with the vaccine used to reduce the chance of it returning. That is the setting where this approach makes the most sense today.

And a design is not a treatment. A licensed veterinarian has to decide whether it is appropriate for your dog, administer it, and monitor them.

Where RosieVaccine fits in

We do the middle of that pipeline. We take a tumor sample and a matched blood sample, identify the mutations unique to your dog's cancer, predict which of them their immune system can see, and design a codon-optimised mRNA blueprint targeting those neoantigens. The vaccine itself is made by a manufacturing partner, and it is shipped to your veterinary oncologist, who administers it and stays in charge of your dog's care throughout.

We are in closed beta and take a small number of pilot cases per quarter. If your dog has been diagnosed with cancer, the right next step is not to contact us. It is to talk to your veterinary oncologist, who can request a pilot case through the form on our home page if they think it is a reasonable fit. And whatever you decide, make sure the tumor sample is preserved, because everything above depends on it.