Cambridge AI Pan-Sarbeco Vaccine Clears First Phase 1 Trial

What Happened

A first-in-human clinical trial of an AI-designed universal coronavirus vaccine has cleared its safety hurdle in the United Kingdom, marking the first time a vaccine whose active ingredient was generated entirely by computer simulations has been dosed into volunteers. Researchers from the University of Cambridge and its spin-out company DIOSynVax announced on June 5 that the experimental shot — known by its development code pEVAC-PS — was safe and well tolerated, with no serious side effects reported across 39 healthy adults aged 18 to 50. The Phase 1 results were published in the Journal of Infection and described as a milestone for a new class of "future-proof" vaccines.

The trial was an open-label, dose-escalation study run at National Institute for Health and Care Research (NIHR) Clinical Research Facilities in Southampton and Cambridge, with University Hospital Southampton NHS Foundation Trust as sponsor. Volunteers had previously received two or three doses of authorized COVID-19 vaccines, allowing the team to measure how the new candidate broadened existing immunity. Rather than a syringe, the team administered the shot as a DNA vaccine through a micro fluid jet — a needle-free system intended to ease vaccination campaigns in clinics where injections are logistically difficult or anxiety-inducing for patients.

Critically, the antigen at the heart of pEVAC-PS was not picked from a single circulating strain. It was built from scratch by a machine-learning pipeline trained on the genetic sequences of multiple Sarbeco coronaviruses. The system identified structural features shared across the whole family — including SARS-CoV-2, the original SARS virus, and several bat coronaviruses flagged in global surveillance programs — and stitched them into what the team calls a "super-antigen." The resulting molecule does not exist in nature; it was engineered to teach the human immune system patterns that are conserved across many viruses at once.

Why It Matters

Most vaccines in widespread use today are built around the version of a virus that is already infecting people. Seasonal flu shots are reformulated every year on the basis of strains circulating in the Southern Hemisphere; the COVID-19 boosters distributed since 2021 have been periodically updated to follow Omicron descendants and successor lineages. Professor Jonathan Heeney of Cambridge, who leads the project, likened that approach to "a dog chasing its tail." A computationally designed super-antigen, by contrast, attempts to anticipate where a virus family can and cannot mutate, then trains the immune system on the parts that stay constant.

The choice of the Sarbeco subgenus is also strategic. This grouping of coronaviruses includes SARS-CoV-1, SARS-CoV-2 and dozens of relatives still circulating in bats across South and East Asia, and zoonotic spillover events from this clade have already triggered two human outbreaks in the past quarter-century. By immunizing volunteers against features shared across the whole Sarbeco family, the Cambridge team is testing whether one shot can broaden protection to viruses that have not yet emerged in humans. The trial showed that recipients produced immune responses not only to SARS-CoV-2 and SARS, but also to several bat coronaviruses that researchers monitor as candidates for the next pandemic.

Beyond Sarbeco viruses, the wider implication is that a single platform — train an AI on a viral family, generate a super-antigen, validate in animals, then test in humans — could be rerun for influenza, Ebola or hemorrhagic fever viruses. DIOSynVax already lists candidates in those areas in its development pipeline. If the approach holds up at scale, public-health agencies may no longer need to wait for a virus to jump species before designing a countermeasure. The shift would be similar to the move from reactive antivirus signatures to behavior-based threat models in cybersecurity: less specific, but more durable.

Reaction

Within the research community the response has been measured optimism. Professor Saul Faust, chief investigator at the University of Southampton, told reporters that the current "reactive" vaccine system "struggles to keep pace" with viral evolution and described the new class as "future-proofed," noting that broad activity against not-yet-emerged relatives could mean "millions of lives could be saved, lockdowns avoided and the economy preserved." Professor Marian Knight, scientific director for NIHR Infrastructure, called the milestone a "pivotal leap forward" enabled by the partnership between life-sciences companies and NHS clinical research sites in Cambridge and Southampton.

Outside the trial team, virologists and vaccine experts have framed the result more cautiously. The Phase 1 study was small — 39 participants, all young and healthy — and was designed to test safety and basic immune response, not real-world protection. Several public-health commentators have noted that the more interesting question is whether the AI-generated super-antigen can produce durable, broadly neutralizing antibodies once trialed in older adults and people with weaker immune systems. Coverage in outlets including pharmaphorum and News-Medical emphasized that comparative immunogenicity data, head-to-head against existing boosters, will be critical to convince regulators.

What's Next

The team is now preparing a larger Phase 2 study to assess the vaccine's ability to induce strong, broadly protective immune responses in a wider and more diverse population. That trial will likely include older adults and participants with varying COVID-19 exposure histories, and it will run alongside continued laboratory work mapping antibody activity against bat coronaviruses that have not yet been observed in humans. The funder, Innovate UK, supplied most of the early-stage budget; further development is expected to involve external biotech partners as DIOSynVax pushes the platform through the regulatory pipeline.

Beyond pEVAC-PS itself, the Cambridge team has signaled that the same digital-design playbook will be applied to additional families. DIOSynVax's reported pipeline already includes candidates aimed at seasonal influenza, pandemic influenza, hemorrhagic fevers and broader coronavirus targets. Independent of DIOSynVax, other groups — including academic labs in the United States and computational-vaccine startups backed by the Wellcome Trust — are exploring similar AI-led approaches. The Cambridge result is the first one to clear human safety testing, but the field is now crowded enough that a second or third candidate is likely to follow within a year, including for influenza, where the World Health Organization has long called for a true universal vaccine.

Closing Thoughts

It is tempting to read the pEVAC-PS trial as a story about software replacing biology, but the more accurate framing is the opposite. The vaccine was not designed by an algorithm acting alone. Computational immunologists fed the system years of accumulated structural, genetic and surveillance data on coronaviruses; veterinary virologists validated the candidate in animals; clinical researchers ran the trial through NHS infrastructure; and regulators imposed every constraint that any other vaccine candidate would face. What is new is not the absence of human judgment, but the speed at which a hypothesis about a virus family can be turned into an antigen ready for testing in a person.

The deeper question raised by Cambridge's announcement is whether public-health systems are ready for a world in which vaccine candidates can be generated faster than they can be evaluated. Phase 1 safety is the cheapest gate in the vaccine pipeline; manufacturing, distribution, immunization records and public trust are not. If AI-designed shots become routine, the bottleneck will move from the lab to the clinic, the cold chain and the conversation between health workers and patients. That is a hopeful kind of problem to have — but it is still a problem that institutions, not models, will need to solve.

한글 요약

케임브리지 대학교와 스핀오프 기업 DIOSynVax가 인공지능으로 설계한 범용 코로나바이러스 백신 후보 pEVAC-PS의 첫 임상 1상 결과를 6월 5일 발표했다. 18~50세 건강한 성인 39명을 대상으로 한 시험에서 백신은 안전하고 내약성이 우수했으며, 주삿바늘 대신 마이크로 플루이드 제트를 이용한 DNA 백신 형태로 투여됐다. 결과는 Journal of Infection에 게재됐고, 자금은 영국 정부 기관 Innovate UK가 지원했다.

핵심은 항원을 사람이 직접 고른 것이 아니라, 머신러닝이 SARS-CoV-2, SARS-CoV-1, 그리고 박쥐 코로나바이러스 등 Sarbeco 코로나바이러스 가족 전체의 유전 정보를 학습해 공통 구조를 모은 "슈퍼 항원"을 생성했다는 점이다. 자연계에 존재하지 않는 이 항원은 변이가 일어나도 변하지 않는 영역을 면역계에 가르치도록 설계됐고, 시험 참가자들에게서 SARS-CoV-2와 SARS는 물론, 아직 인간에게 감염되지 않은 박쥐 바이러스에 대한 면역 반응까지 끌어냈다.

다음 단계는 노년층과 면역 상태가 다양한 더 넓은 모집단을 대상으로 한 임상 2상이며, DIOSynVax는 같은 플랫폼을 인플루엔자와 에볼라 등 다른 바이러스 가족에도 확장할 계획이다. 인공지능이 백신 설계 속도를 끌어올린 만큼, 앞으로는 실험실보다 임상 평가, 제조, 유통, 그리고 보건의료 현장에서의 신뢰 구축이 새로운 병목이 될 가능성이 크다.