How a machine-learning-guided personal cancer vaccine kept nine kidney cancer patients recurrence-free for more than three years, and why Frontier scores it 81 rather than 100.
Nine out of nine patients with high-risk, surgically removed kidney cancer were still cancer-free at a median of 40.2 months after surgery, every one of them having mounted a measurable immune response to a vaccine built from the private mutations of their own tumor. That single sentence is why the phrase "AI-designed personal cancer vaccine reports ~100% cancer-free at 3 years" traveled across feeds in 2026, and it is, as stated, true. It is also one of the most easily misread results in modern oncology, because almost everything that makes it exciting (a bespoke vaccine, a perfect recurrence record, a durable immune response) is also exactly the kind of thing that a small early trial can produce by a mix of real biology and favorable circumstance.
The result is genuine, and it belongs to a class of medicine that had, by late 2026, finally produced its first hard proof. But "100 percent" is not a cure rate, "AI-designed" is a compressed description of something more specific, and "three years" is a median, not a guarantee. This guide takes the breakthrough apart from first principles: what the trial actually did, where the artificial intelligence really sits in the pipeline, why the number nine matters as much as the number one hundred, and how a nine-patient study with no control arm sits inside a field that, in the same season, delivered a 1,137-patient phase 3 win and a quietly terminated colorectal trial. This is the Frontier lens applied to a single result: evidence, impact, and novelty, each weighed separately, with every figure traceable to its primary source.
Contents
- The result, stated precisely
- Reading it through Frontier's lens: why 81, not 100
- What a neoantigen vaccine actually is
- Where the "AI" really sits, and where it does not
- Why "100 percent" is the most dangerous number in the story
- The class grows up: the melanoma phase 3 that changes the argument
- The counter-evidence: pancreatic persistence and a colorectal failure
- Manufacturing, cost, and the access problem nobody solved yet
- What would move the score from signal to proof
- Conclusion: how to hold a result like this
1. The result, stated precisely
The breakthrough is a phase 1 clinical trial of a personalized neoantigen vaccine in patients with high-risk, fully resected clear cell renal cell carcinoma, the most common form of kidney cancer. It was published in Nature in February 2025 under the title "A neoantigen vaccine generates antitumour immunity in renal cell carcinoma" - Nature. The trial (ClinicalTrials.gov identifier NCT02950766) enrolled just nine participants with stage III or IV disease who had undergone surgery with curative intent, recruiting them between March 2019 and September 2021 - Journal of Clinical Oncology, via search records. Every patient's tumor was sequenced, its unique mutations catalogued, and a vaccine assembled to match.
The headline outcome is the recurrence record. At a median follow-up of 40.2 months after surgery, none of the nine had experienced a recurrence of their cancer - CancerNetwork. Measured instead from the first dose of vaccine, the median follow-up was 34.7 months, and the median disease-free survival had not been reached, meaning more than half the cohort had not recurred by the time the data were locked - Targeted Oncology. The National Cancer Institute, summarizing the work, noted the granular version: no patient had recurred, "including four who were still cancer-free more than 3 years after their surgery" - National Cancer Institute. That is the honest shape of the "cancer-free at 3 years" claim: a small cohort, a median follow-up just past three years, and a clean recurrence record within it.
The immunology underneath the recurrence record is arguably the stronger part of the paper. Every one of the nine patients generated a T cell response against the vaccine's targets, and the response was fast and large: it appeared within about three weeks of the first dose and expanded to a mean 166-fold increase in vaccine-induced T cells, with high levels of those cells still detectable up to three years later - Dana-Farber Cancer Institute. Critically, the immune system was taught to recognize the mutations that drive kidney cancer itself, including alterations in the genes VHL, PBRM1, BAP1, KDM5C and PIK3CA - Urology Times. A vaccine that provokes durable immunity against the tumor's own engine, rather than against an incidental passenger mutation, is a more convincing mechanism than the recurrence count alone.
The vaccine here is a synthetic long-peptide product, not the messenger RNA format that the phrase "cancer vaccine" now tends to summon. Each patient received four pools of up to five synthetic long peptides, so up to 20 neoantigens in total, dosed at 300 micrograms per peptide and mixed with the immune adjuvant poly-ICLC at half a milligram per pool, given under the skin and into the skin across seven visits over 24 weeks - Journal of Clinical Oncology, via search records. Five of the nine patients also received the checkpoint inhibitor ipilimumab alongside the vaccine, while four received the vaccine alone, a design detail that matters for interpretation later. There were no dose-limiting toxicities; the most common side effects were low-grade injection-site reactions in all nine patients and transient flu-like symptoms in eight - Targeted Oncology.
The work came from Dana-Farber Cancer Institute, the Broad Institute of MIT and Harvard, and Yale Cancer Center, with the lead author, David Braun, now at Yale - Dana-Farber Cancer Institute. Braun framed the finding in terms of what was learned rather than what was proven: "We learned which specific targets in the cancer are most susceptible to immune attack and demonstrated that this approach can generate long-lasting immune responses" - National Cancer Institute. That is the correct register for a phase 1 result, and it is the register the rest of this guide will try to hold.
2. Reading it through Frontier's lens: why 81, not 100
Frontier scores every breakthrough on three separate axes and refuses to blend them into a single verdict prematurely. The Evidence pillar asks how well established a result is: the size and design of the study, whether it is corroborated by independent work, whether it cleared peer review. The Impact pillar asks how much it could matter: the size of the population it could help and the momentum it has already gathered. The Novelty pillar asks how genuinely new it is: whether it opens a mechanism nobody had shown, or applies a known method to a new target. A result can be high on one axis and low on another, and the point of keeping them apart is to stop a moving story from borrowing credibility it has not earned. For the full method, see the latest medical breakthroughs of 2026, scored, which walks through the same machinery across 27 results.
This breakthrough carries a Frontier Score of 81, which is high for the index and demands an explanation, because on evidence alone a nine-patient uncontrolled trial should not score that well. The score is not a statement that the kidney vaccine is proven. It is a composite that reflects three things at once: the unusual internal strength of this particular phase 1 (a clean recurrence record paired with a documented, durable, on-target immune response in every patient), the corroboration from the wider vaccine class that arrived in 2026 (a positive phase 3 in melanoma and multi-year persistence data in pancreatic cancer), and the impact of the target (an adjuvant strategy for a cancer where the standard of care still leaves a large fraction of patients to recur). A number in the low 80s says the signal is strong and increasingly corroborated across the field, not that the case is closed.
The reason it is 81 and not higher is the same reason a careful reader should not say "cure." The trial has no control arm, nine patients, and a follow-up that has only just crossed the three-year median. Half of these patients also received a checkpoint inhibitor, which independently reduces recurrence risk, so the vaccine's specific contribution cannot be cleanly separated from ipilimumab's in a study this small. Those are not footnotes; in oncology they are the difference between a promise and a proof, and Frontier's scoring is calibrated to them. The gap between 81 and 100 is precisely the gap between "a strong, corroborated signal" and "a randomized demonstration that the vaccine caused the benefit." That gap is not a criticism of the work. It is the work that remains.
3. What a neoantigen vaccine actually is
To judge the result you have to understand the target, and the target is a neoantigen. When a cell becomes cancerous it accumulates mutations, and some of those mutations change the amino acid sequence of the proteins the cell makes. Fragments of those altered proteins are displayed on the cell surface by molecules called human leukocyte antigens (HLA), the same display system the immune system uses to distinguish self from foreign. A neoantigen is one of these mutated fragments: a piece of protein that exists only in the tumor and nowhere in the patient's healthy tissue. Because it is genuinely foreign to the body, it is a target the immune system can attack without attacking normal cells, which is the central promise of the whole approach.
The problem is that neoantigens are, by definition, unique to each tumor. There is no off-the-shelf neoantigen the way there is an off-the-shelf spike protein for a virus. A personalized cancer vaccine therefore has to be built from scratch for every patient: sequence the tumor, find the mutations, work out which mutated fragments will actually be displayed on that specific patient's HLA molecules, choose the most promising ones, and manufacture a vaccine that presents them to the immune system in a form it will respond to. The vaccine does not kill the cancer directly. It trains T cells to recognize the tumor's private mutations, so that any cancer cell left behind after surgery, or any that returns, is met by an immune system that already knows what to look for. That is why these vaccines are given in the adjuvant setting, after surgery has removed the visible disease, as a way to prevent recurrence rather than to shrink an existing mass.
The two dominant formats deliver the same idea through different chemistry. The kidney cancer vaccine used synthetic long peptides: the neoantigen fragments are chemically synthesized and injected with an adjuvant that alerts the immune system. The melanoma and pancreatic programs discussed later use messenger RNA, encapsulated in a lipid nanoparticle, that instructs the patient's own cells to manufacture the neoantigens internally. The mRNA format is the one made famous by COVID vaccines, and it carries a subtle engineering trick: the RNA is chemically modified with pseudouridine so that, in the words of one National Cancer Institute researcher, "the modification seems to make the mRNA itself almost invisible to the immune system," letting the encoded neoantigens rather than the delivery vehicle draw the immune response - National Cancer Institute. The mRNA format also scales to more targets: an mRNA vaccine can encode up to 34 neoantigens in a single construct, against roughly 20 for the peptide approach.
Neither format works alone, and this is the part most summaries omit. Personalized vaccines are almost always paired with a checkpoint inhibitor, a drug that releases the brakes tumors use to shut down T cells. The vaccine supplies the target; the checkpoint drug keeps the responding T cells from being switched off. In the kidney trial, five of nine patients received ipilimumab; in melanoma and pancreatic cancer the vaccines are given with pembrolizumab and atezolizumab respectively. The mental model is not "a vaccine that cures cancer" but "a vaccine that points an already-unleashed immune system at the right target." That combination is why the field is only now, in 2025 and 2026, producing durable results: the checkpoint drugs it depends on became standard of care only in the last decade.
4. Where the "AI" really sits, and where it does not
The label "AI-designed" is doing a lot of compression, and it is worth decompressing carefully, because the honest version is both less magical and more interesting than the headline. No generative model wrote a vaccine sequence from nothing here. What machine learning does in this pipeline is prediction and prioritization: given the hundreds of mutations found in a tumor, and the specific HLA molecules a patient carries, which handful of mutated fragments will actually be displayed on the cell surface, and which of those displayed fragments will provoke a T cell response? That is a filtering problem across a combinatorial space no human can search by hand, and it is where the algorithms earn their place.
The workhorses are peptide-HLA binding predictors, and the best known is NetMHCpan, a pan-specific artificial neural network that predicts how strongly a given peptide will bind a given HLA molecule - Nucleic Acids Research. Successive versions have folded in more data and better methods: version 4.1 integrated mass-spectrometry measurements of peptides actually eluted from HLA molecules, and a 2025 release added transfer learning and structural features to improve the prediction of the specific epitopes that CD8 T cells see - Frontiers in Immunology. Around these sit related models such as MHCflurry and MixMHCpred, and a growing wave of deep-learning approaches trained directly on immunopeptidomics data - PMC. Dana-Farber's own pipeline, which supported this trial and several others, uses immunopeptidome-trained epitope prediction to design the long peptides and to prioritize highly expressed neoantigens and those arising from cancer drivers - Journal of Immunology. This is the sense in which the vaccine is "AI-designed": machine learning chooses the targets.
The honest caveat is that these predictors are useful but far from perfect, and their imperfection is the field's rate-limiting step. The National Cancer Institute's plain-language description is accurate: "Computer algorithms then predict which neoantigens are most likely to bind to receptors on T cells and stimulate an immune response" - National Cancer Institute. The word "predict" is load-bearing. Reviews of the area are candid that high false-positive rates, limited diversity in training data, and the sheer biological complexity of immunogenicity mean that many predicted neoantigens never provoke a response, and some genuine ones are missed - Frontiers in Genetics. A vaccine that targets up to 20 or 34 fragments is, in part, a hedge against the fact that the model cannot yet tell you which two or three will do the real work.
This is a familiar pattern for anyone tracking artificial intelligence as it moves from demonstrations into working science: the value is not a machine replacing the researcher but a machine compressing a search that was previously intractable, with a human-and-experiment loop still required to confirm what the model proposes. The same shape shows up wherever AI meets a hard scientific instrument, from protein design to the way a network can reconstruct the structure of Earth's core from seismic data. In cancer vaccines the loop is: predict, manufacture, vaccinate, measure the actual T cell response, and feed that back. The kidney trial's finding that all nine patients mounted a response, much of it against driver mutations, is partly a validation that the prediction step is good enough to be useful, which is a real and underappreciated result inside the flashier recurrence number.
5. Why "100 percent" is the most dangerous number in the story
A perfect recurrence record in nine patients is exactly the kind of statistic that is both true and misleading, and the most important corrective comes from an oncologist with no stake in the trial. Mark Ball of the National Cancer Institute, commenting on the kidney result, made the point that undoes the naive reading: "it's not unusual for people with operable kidney cancer to live for several years or longer after surgery without their cancer coming back" - National Cancer Institute. Some fraction of these nine patients would very likely have stayed recurrence-free on surgery alone. With nine patients and no control group, there is no way to know how many, which is why the immune data (a documented, on-target, durable T cell response in every patient) is the more compelling half of the paper than the survival curve.
The scale of the interpretive trap becomes clear when you look at what "high-risk" recurrence actually means in this disease, using the best randomized data available. The KEYNOTE-564 trial tested adjuvant pembrolizumab against placebo in high-risk clear cell kidney cancer, the current standard of care. At five years, the estimated recurrence-free rate was 63.3 percent with pembrolizumab versus 54.5 percent with placebo - Journal of Clinical Oncology. Read those numbers carefully. Even with the best approved adjuvant therapy, roughly 37 percent of high-risk patients recur within five years, and even on placebo, a majority (about 55 percent) do not. That is the backdrop against which nine-for-nine has to be judged: impressive, but drawn from a distribution where clean multi-year runs are not rare, and measured over a follow-up that has only just reached the point where recurrences cluster.
None of this makes the vaccine result unimportant. It makes it early, which is a different thing. The correct inference from a clean phase 1 is not "this works" but "this is safe, it does what it was designed to do immunologically, and it is worth testing in a design that can actually prove efficacy." The investigators say as much, and independent commentators reinforce it: the sample is small, there is no control arm, and larger confirmatory trials are needed before anyone can attribute the recurrence record to the vaccine - National Cancer Institute. This is the same discipline Frontier applies across its index, and it is why the biggest scientific breakthroughs of 2026 so often score below where their headlines would place them. A result can be real, safe, mechanistically convincing, and still not yet be proof of benefit. All of those things are true here at once.
6. The class grows up: the melanoma phase 3 that changes the argument
The reason a nine-patient kidney trial can carry a Frontier Score of 81 is that it no longer stands alone. In August 2026, the same broad idea (a personalized neoantigen vaccine given after surgery with a checkpoint inhibitor) produced its first positive phase 3 result, and it did so at a scale that answers the objections raised above. Merck and Moderna announced that the phase 3 INTerpath-001 trial (NCT05933577) met its primary endpoint of recurrence-free survival and a key secondary endpoint of distant metastasis-free survival in patients with completely resected stage IIB to IV melanoma - Merck. The vaccine, now named intismeran autogene (previously mRNA-4157 or V940), is an mRNA construct encoding up to 34 neoantigens designed from each patient's tumor, given with pembrolizumab.
The scale is the point. INTerpath-001 enrolled 1,137 patients and compared the vaccine plus pembrolizumab against pembrolizumab alone, the actual standard of care, in a randomized design - Merck. This is the design the kidney trial lacked, at a hundred times the patient count, and it is why the ASCO Post described the readout as the first for any individualized neoantigen therapy and the first for an mRNA-based cancer therapy to clear phase 3 - The ASCO Post. The comparison arm matters enormously: beating placebo is one thing, but beating pembrolizumab, a drug that already prevents a large share of recurrences, is a far higher bar, and it is the bar this trial cleared - ecancer.
The phase 3 win did not come from nowhere; it was foreshadowed by the phase 2b KEYNOTE-942 trial, whose results were published in the Lancet and matured over several years - The Lancet. In that earlier study, the vaccine plus pembrolizumab produced an 18-month recurrence-free survival of about 79 percent versus 62 percent for pembrolizumab alone, an early separation that held up as the data matured - via search records, CancerNetwork.
Expressed as risk reduction, the same phase 2b data showed the combination cutting the risk of recurrence or death by 49 percent (hazard ratio 0.51) and the risk of distant metastasis or death by 59 percent (hazard ratio 0.411) compared with pembrolizumab alone, a benefit that was still being reported at five years of follow-up at the 2026 ASCO meeting - Merck. These are the numbers that made the phase 3 worth running, and their confirmation in a 1,137-patient trial is what retroactively strengthens the interpretation of every smaller study in the class, including the kidney vaccine.
The logical relationship between the melanoma phase 3 and the kidney phase 1 is worth stating explicitly, because it is easy to overstate. The melanoma trial does not prove the kidney vaccine works; they are different cancers, different vaccine formats (mRNA versus peptide), and different checkpoint partners. What the melanoma trial proves is that the general strategy (personalize a neoantigen vaccine, pair it with a checkpoint inhibitor, give it after surgery) can produce a statistically robust reduction in recurrence in a large randomized trial. That raises the prior probability that the same strategy, applied to kidney cancer, is doing something real rather than riding luck. It is corroboration of the class, not proof of the instance, and the Frontier Score reflects exactly that distinction.
7. The counter-evidence: pancreatic persistence and a colorectal failure
A serious reading of any breakthrough requires looking hard for the results that cut the other way, and the vaccine class provides both a reinforcing case and a cautionary one in the same year. The reinforcing case is pancreatic cancer, one of the deadliest and least mutated of all tumors, and therefore one of the hardest possible tests of a mutation-targeting vaccine. In the phase 1 trial of autogene cevumeran, a personalized mRNA vaccine from BioNTech and Genentech, 16 patients with resected pancreatic cancer received the vaccine with the checkpoint inhibitor atezolizumab and chemotherapy; 8 of the 16 mounted a strong neoantigen-specific T cell response - Nature, via search records. The durability data are what elevated the result: at multi-year follow-up, the patients who responded had recurrence-free survival that had still not reached its median, against 13.4 months in the non-responders, and the vaccine-induced T cell clones were estimated to persist for years - BioNTech. Frontier tracks this result in its index as a personalized mRNA vaccine that builds years-long immunity against pancreatic cancer, and a randomized phase 2 (IMcode003, NCT05968326) is now underway to test whether that persistence translates into a survival benefit.
The same persistence signature has now shown up in a third cancer, which is part of why the class as a whole is gaining credibility rather than resting on one tumor type. In triple-negative breast cancer, an individualized mRNA vaccine given after surgery produced high-magnitude, mostly new T cell responses that stayed functional for years, a result Frontier indexes as individualized mRNA vaccines that evoke durable T cell immunity. Across kidney, melanoma, pancreatic, and breast cancer, the recurring finding is the same: in the patients who respond, the immunity lasts. That consistency across independent teams and tumor types is a stronger form of evidence than any single trial, and it is the real reason the kidney result reads as a signal rather than a fluke.
But the cautionary case is essential, and it is the part the celebratory coverage skips. In August 2026, BioNTech confirmed that a phase 2 trial of the very same autogene cevumeran vaccine, given as an adjuvant monotherapy in patients with circulating-tumor-DNA-positive, resected stage II or III colorectal cancer (the BNT122-01 trial, NCT04486378), had been terminated on the recommendation of its independent data safety monitoring board, which identified a numerical imbalance in overall survival between the arms and concluded that continuing was unlikely to change the outcome - BioNTech. No new safety signal was found, and the company was careful to note that its pancreatic program was unaffected. The instructive detail is in the design: the colorectal trial tested the vaccine as monotherapy, without a checkpoint inhibitor, which is exactly the pairing the mechanism predicts you need. A vaccine points the immune system at a target; without a checkpoint drug to keep the responding T cells switched on, the strategy may simply not have enough to work with.
The colorectal failure is not a reason to dismiss the class, but it is a decisive reason to resist the "cancer vaccines cure cancer" framing. It demonstrates that the same technology can miss, that the choice of combination and setting is not a detail but the whole ballgame, and that a personalized neoantigen vaccine is a platform whose results depend on how it is deployed, not a single therapy with a single verdict. Held next to the kidney and melanoma wins, it draws the real boundary of the 2026 evidence: this works best as an adjuvant, in combination with a checkpoint inhibitor, in cancers where the immune system can be productively engaged, and it should not be assumed to work outside those conditions until a trial says so.
8. Manufacturing, cost, and the access problem nobody solved yet
Even granting that the biology is real and increasingly proven, a personalized vaccine faces an obstacle that an off-the-shelf drug does not: every dose is a manufactured-to-order product with a supply chain of one. The workflow is genuinely bespoke. After the tumor is removed it must be sequenced (one to two weeks), the neoantigens predicted and prioritized (about a week), and the individualized vaccine synthesized and quality-checked (two to four weeks), for a typical turnaround of roughly four to seven weeks from surgery to first dose - ScienceDirect. Moderna has described making each personalized melanoma vaccine in about six weeks - National Cancer Institute. For an adjuvant therapy given after surgery, that lag is tolerable; for treating active, growing disease it is a much harder constraint.
The economics follow the logistics. A conventional drug amortizes an enormous fixed development cost across many identical doses, which is what makes it affordable per patient. A personalized vaccine cannot do this in the same way, because the manufacturing is repeated in full for every single patient. That is why the near-term commercial models concentrate on adjuvant settings in higher-value indications, where preventing a recurrence justifies the per-patient manufacturing cost, and why the melanoma program (a common cancer with a clear high-risk resected population and an established checkpoint backbone) was the rational place for the first phase 3. Scaling to the long tail of cancers will require driving the cost and time of the design-and-manufacture loop down, which is precisely where better machine-learning prediction and more automated synthesis are expected to contribute over the next few years - PMC review.
There is also a structural fragility that has nothing to do with the science: much of this work depends on public research funding at the discovery and early-trial stage, and that funding is not guaranteed. Coverage in 2026 noted explicitly that personalized mRNA vaccines are close to clinical reality but that cuts to federal research budgets could slow or derail the pipeline that feeds them - Scientific American. The kidney trial itself was funded by a mix of a cancer research foundation, the US Department of Defense, and the National Institutes of Health - Dana-Farber Cancer Institute. The point for a reader trying to gauge how fast this reaches patients is that the binding constraints are now at least as much manufacturing, cost, and policy as they are biological, and progress on the science does not automatically dissolve them.
9. What would move the score from signal to proof
Frontier scores are not static; they are meant to move as evidence accumulates, and it is worth being concrete about what would push the kidney cancer vaccine's 81 up or down. The single most important thing is a randomized, controlled trial in kidney cancer specifically, large enough to separate the vaccine's effect from the checkpoint inhibitor's and from the natural history of the disease. The phase 1 established safety and immunogenicity; only a randomized trial with a proper comparison arm can establish that the vaccine causes fewer recurrences. Until that exists for kidney cancer, the score reflects a strong and corroborated signal, not a demonstrated benefit in this disease.
The evidence a result like this has to climb is a ladder, and it is useful to see the rungs, because most public discussion collapses them into a single "it works" or "it doesn't." A clean phase 1 sits near the bottom of the ladder, not the top, however good its numbers look.
Several concrete readouts over the next few years will do most of the work of resolving the picture, and they are worth watching as a set rather than individually. The melanoma phase 3 needs to convert its recurrence-free and metastasis-free wins into an overall survival benefit, the endpoint regulators and patients care about most, and that data is still maturing - Merck. The randomized pancreatic phase 2 needs to show that the striking phase 1 persistence translates into fewer recurrences when tested against a control. And kidney cancer needs its own randomized trial to begin. If those readouts are positive, the class moves decisively from "promising" to "standard of care in the making," and the individual scores rise with it. If the overall survival data disappoint, or the pancreatic randomization fails to reproduce the phase 1 signal, the whole class is repriced downward, and the colorectal termination stops looking like an outlier.
The deeper scientific question, the one that will determine how far this generalizes, is why some patients respond and others do not. In every trial in the class, the benefit concentrates in the patients who mount a strong T cell response, and a meaningful fraction do not respond at all. Improving that response rate is partly a prediction problem (choosing better targets, which is where the machine-learning models have the most room to improve) and partly a biology problem (understanding which tumors are immunologically "cold" and cannot be productively vaccinated regardless of target choice). Reviews of the field are explicit that clinical benefit remains inconsistent for exactly these reasons - ScienceDirect. The results that would most move Frontier's assessment of the whole approach are the ones that turn non-responders into responders, because that is the difference between a therapy for the lucky subset and a therapy for the disease.
10. Conclusion: how to hold a result like this
The right way to hold "AI-designed personal cancer vaccine reports ~100% cancer-free at 3 years" is to hold three separate facts at once, without letting any of them swallow the others. The first is that the underlying result is real and genuinely encouraging: nine high-risk kidney cancer patients, all recurrence-free at a median of over three years, all with a documented, durable, on-target immune response, and no serious safety problems. The second is that the result is early and uncontrolled: nine patients, no comparison arm, half of them also on a checkpoint inhibitor, and a follow-up that has only just reached the point where recurrences matter most. The third is that the result no longer stands alone: a 1,137-patient phase 3 in melanoma cleared the bar in 2026, multi-year persistence has now been shown across pancreatic and breast cancer, and a colorectal trial failed in a way that maps cleanly onto the mechanism.
Put together, those three facts are why Frontier scores this breakthrough 81: high enough to say the signal is strong and increasingly corroborated across the vaccine class, short of the range reserved for results proven in randomized trials. For a reader deciding how much weight to put on the news, the decision framework is straightforward. Treat the kidney result as a strong hypothesis with a good mechanism and growing external support, not as a cure and not as a reason to expect a kidney cancer vaccine in the clinic tomorrow. Watch the overall survival data from melanoma and the randomized readout from pancreatic cancer as the events that will confirm or complicate the story. And keep the colorectal failure in view as the reminder that this is a platform whose results depend entirely on how it is deployed, not a single therapy with a single verdict.
The larger significance is that 2025 and 2026 are the years the personalized cancer vaccine stopped being a perennial "five years away" and started producing the specific kind of evidence that changes practice: durable immunity documented over years, and a randomized phase 3 that beat the standard of care rather than a placebo. The kidney trial is one of the cleaner early entries in that shift, and its imperfections are not a mark against it but a precise description of where the work now stands. For the full scored landscape it sits inside, from in-body gene editing to AI-designed proteins, the latest medical breakthroughs of 2026 map the same terrain result by result, each held to the same standard: real data, honest caveats, and one traceable number.
This guide reflects the evidence as of September 2026. Clinical results, trial statuses, and regulatory decisions change; the figures here are drawn from the primary trial reports and peer-reviewed publications linked throughout, and readers should verify current status before drawing conclusions about any specific therapy.