Developing new vaccines is a long and costly process, and researchers need strong evidence to confirm the efficacy of a vaccine candidate before it can achieve regulatory approval. Correlates of protection (CoP)—immune biomarkers that signal whether someone is protected from a disease—can help speed the vaccine development process by more rapidly and reliably determining whether new candidates will be effective.
Using CoPs can shorten the often long—sometimes 10 or more years—journey of vaccine candidates toward regulatory approval, getting safe and protective vaccines to the children and communities that need them sooner. Researchers were able to update COVID-19 booster shots more quickly, for instance, because they determined that a simple antibody blood test was a reliable indicator of protection—eliminating the need for lengthy clinical trials of the updated vaccine.
PATH is investigating how agentic AI workflows—in which multiple AI agents can autonomously coordinate, plan, and take actions to achieve a defined goal—can complement scientific expertise to identify potential immune biomarkers. This use of AI, if successful, could transform vaccine development, leading to more efficient evaluation of new candidates and reducing the time and resources needed to bring lifesaving vaccines to market.
Tapping into decades of data
Discovering and validating potential immune biomarkers is a highly complex and time-consuming process. These biomarkers can be identified in several ways, including studies of recovery from natural infection, animal immunogenicity studies, and human clinical studies. Scientific literature related to the development of vaccines against a given pathogen can cover decades of research and be highly diverse.
AI offers a powerful opportunity for accelerating this process. By rapidly analyzing data, generating hypotheses, and supporting regulatory justifications, AI tools may be able to identify potential CoPs that could help speed the development of new vaccines.
“Automated search and synthesis of research is not new, but as with so many fields, the advent of large language models has profoundly changed what is possible—and the speed at which it can be done,” said Dr. Bilal Mateen, Chief AI Officer at PATH.
“These tools have exciting potential to not only breathe a second life back into previously considered but prematurely abandoned hypotheses, but also, given their 'generative' nature, to maybe even propose something radically new.”
A co-scientist for CoP researchers
PATH is testing the ability of an “AI co-scientist”—an AI tool built for hypothesis generation and research tasks—to identify potential immune biomarkers for diseases including rotavirus and respiratory syncytial virus (RSV). Experts will carefully review these AI-generated hypotheses and test them to assess their scientific validity.
“As co-scientist tools evolve, we’re identifying ways we can enhance and tailor these models for CoP discovery.”— Dr. Roberto Amato, Deputy Director of AI4Science, PATH
PATH is also exploring how to improve existing AI tools to better support immune biomarker research. This may involve augmenting existing AI co-scientist tools or building new ones to perform specific tasks, such as statistical analysis.
“The landscape of AI tools is constantly expanding,” said Dr. Roberto Amato, Deputy Director of AI4Science at PATH. “As co-scientist tools evolve, we’re identifying ways we can enhance and tailor these models for CoP discovery, unlocking insights that could accelerate progress on developing new vaccines to address pressing global health challenges.”
Leveraging AI for vaccine development
In addition to identifying potential immune biomarkers, AI co-scientists may be able to support the vaccine authorization process itself. PATH is evaluating the ability of the co-scientist—and other similar tools—to generate clear, evidence-based rationales to support the use of potential CoPs in regulatory submissions.
PATH aims to build a stronger foundation for AI applications in CoP research, catalyzing more efficient development and approval of new vaccines—with the ultimate goal of bringing lifesaving vaccines to market sooner.
“Identifying correlates of protection can shorten the runway from vaccine discovery to delivery,” Dr. Amato said. “We’re putting new AI tools to the test to see how these models can help us get there.”
