LONDON, Oct. 16 (Xinhua) -- British researchers are bringing artificial intelligence (AI), automated laboratories and supercomputing together to accelerate the development of personalized cancer vaccines. The effort aims to test whether a more connected research process can shorten the journey from identifying a tumor target to treating a patient.
The University of Oxford announced early this month that the Advanced Research and Invention Agency (ARIA) was providing an additional 8 million pounds (about 10.6 million U.S. dollars) for the university's UK Cancer Vaccine AI Scientist and Supercomputing Project. The project combines AI tools for selecting vaccine targets with automated laboratory experiments and data analysis, under the guidance of oncologists.
"AI gives us an extraordinary opportunity to find vaccine targets faster and learn from every experiment," said Dr Lennard Lee at Oxford's Center for Immuno-Oncology.
Personalized cancer vaccines aim to direct a patient's immune system against their tumor. A central challenge is identifying tumor fragments that will trigger an effective immune response.
In a January article published in the European Medical Journal (EMJ), researchers including Lee argued that existing computational tools often address separate steps in this process. Current tools can help predict whether a tumor protein fragment will bind to a human leukocyte antigen (HLA) molecule, which helps display it to immune cells. But even strong binding does not necessarily trigger an immune response.
The researchers proposed that AI models trained on genetic information, protein data, tissue images and immune responses could help connect these biological processes and improve the selection of vaccine targets. This remains a research strategy whose benefits require testing, according to the article.
According to Oxford, early laboratory tests during its cancer vaccine project's initial nine-month phase suggested that vaccine targets selected by AI could trigger measurable immune responses. But these findings do not establish that the approach improves survival or controls cancer in patients.
In the next phase, the team plans to develop automated systems to manufacture personalized mRNA cancer vaccines close to patients. A planned Phase I trial will assess the vaccines' safety and feasibility and whether they trigger the intended immune response.
Brian Wang, an ARIA program director, said cancer vaccine discovery was slowed by the separation of ideas, experiments and analysis across different teams. Oxford's project connects AI-generated targets, automated testing and clinical guidance in a single research loop, he added.
DATA AND TRUST CHALLENGES
Progress also depends on access to reliable biological and clinical data that researchers can use to train and test AI models.
A policy report published in June by the Cancer Research UK identifies fragmented datasets and inconsistent interpretations of governance requirements across institutions as obstacles to AI research. Inconsistent interpretations of patient consent, ethics requirements and data-sharing rules can make it harder to combine data across institutions, slowing research. The report calls for clearer governance and data that meet shared standards.
Shivani Sinha, a policy adviser in science and research policy at Cancer Research UK, described the practical consequences in an article: researchers can spend months seeking permission across institutions, only to discover that the data they obtain lack essential information.
"It inherits the gaps in the data and infrastructure around it," Sinha wrote of AI, warning that reuse across projects can reproduce those weaknesses at scale. She argued that datasets must represent the population more adequately and that their limitations must be made clear.
Trust also has a scientific dimension. Researchers need to establish when an AI prediction deserves further investigation and when its apparent precision exceeds the underlying evidence.
In their EMJ article, Lee and colleagues cautioned that AI foundation models can be difficult to interpret and may miss biological interactions important to cancer. They emphasized validation and benchmarking across diverse immune-system profiles, including populations underrepresented in existing datasets.
According to Cancer Research UK's report, scientists need the judgment to assess AI's limitations and recognize when its use is appropriate. Training should develop that critical capacity alongside technical skills.
Public confidence presents a related challenge. The report links trust to accountable governance and meaningful patient involvement in decisions about data access and research oversight. Concerns about privacy, security and unauthorized use of information could undermine the willingness to share data on which these projects depend.
For vaccine development, these challenges are closely connected: incomplete data can weaken predictions, while poorly explained or governed research can make the data needed to improve them harder to obtain.
EXPANDING COMPUTE, ADAPTING OVERSIGHT
Britain is investing public resources to give researchers greater access to the computing capacity needed for this work.
In January, Britain's Department for Science, Innovation and Technology announced an investment of 36 million pounds (about 47.7 million U.S. dollars) to increase the supercomputing capacity available at the University of Cambridge through the national AI Research Resource program to six times its existing level.
The announcement said the national program provides free computing access to eligible researchers and start-ups. Cambridge's DAWN supercomputer had already supported more than 350 projects across different fields, including work on AI tools that could accelerate personalized cancer vaccine development.
Access arrangements remain important. Sinha argued that national capacity figures mean little if researchers cannot secure time on the machines. She also called for bringing analysis to securely held data, reducing the need to move sensitive information between systems.
Regulators face a parallel challenge as AI potentially increases the pace and volume of new treatments.
Lawrence Tallon, chief executive officer of Britain's Medicines and Healthcare products Regulatory Agency (MHRA), told the House of Lords Science and Technology Committee in June that the agency needed greater capacity, specialist expertise and technological investment to handle a growing pipeline.
Discussing AI tools used for medical purposes, Tallon called for closer monitoring after such products reach the market. He explained that an AI system's performance can change as the system learns and evolves. Tallon also described "predetermined change control plans," which set limits in advance on how much an approved software product can change before further regulatory review is required.
The MHRA was also exploring AI to assist its own review of clinical trial applications, while retaining human responsibility, according to Tallon.
"I emphasize that, for the foreseeable future, it will always be a human making the final decision. But AI can speed up our assessment and speed up product development," Tallon said. ■
