A University of York-led study found distinct vaccine-induced immune response biomarkers between HIV-positive and HIV-negative groups, but outliers highlight the diverse and complex nature of the immune system.
Toronto, March. 4, 2026 – How people with compromised immune systems respond to vaccines is an important area of immunology research. A new study led by the University of York found that machine learning models were not only able to accurately identify differences between healthy controls and people infected with HIV, but also found outliers in both groups that offered an interesting glimpse into the complex nature of the immune system and what future personalized medicine might look like, taking into account variables such as age, comorbidities and genetics.
“This study represents an important step forward in the potential of personalized vaccination intervention strategies,” said lead author Chapin Kolosek. He worked on this paper as a postdoctoral fellow at the University of York, under the supervision of Professor Jane Heffernan from the Faculty of Science, and his research focus is on infectious disease modeling. “Learning the structure of immune variation at scale moves us toward a data-driven foundation for personalized vaccination and treatment design.”
Kolosek, now an adjunct professor at the University of Guelph, used a dataset of people with and without HIV who received up to five doses of the COVID-19 vaccine over 100 weeks. All HIV-infected patients were from the Greater Toronto Area and had their disease controlled with antiretroviral therapy. The researchers used a type of machine learning technique called random forests to analyze 64 immune biomarkers triggered by responses to COVID-19 vaccines, creating a group of “virtual patients” to further model immune responses.
“While we were working with a rich dataset suitable for statistical testing, long-term mathematical models still face discriminability limits when the data cannot uniquely resolve immune dynamics. So we turned to machine learning to identify core differences between groups, and then leveraged that learned structure to generate virtual patients that captured how immune patterns differed between groups.”
They were able to show that saliva-based antibodies, specifically a type of antibody found in saliva called IgA, bind to white blood cells long known to be associated with HIV infection, creating characteristic differences between the two groups. Kolosek says this is important because there are many studies showing changes in mucosal immunity in people living with HIV and how it is affected in the short and long term.
Heffernan noted that they identified subgroups within the HIV-positive group, which highlights the importance of personalized vaccination strategies and the challenges of modeling immune responses due to individual differences.
“The immune response is very complex,” Heffernan explains. “Sometimes they act as inhibitors of a group of immune responses, while others are activators. There are large individual differences even among people with similar immune system conditions. Using machine learning, mechanistic modeling, and ‘virtual patients’ we can uncover important differences within subgroups and between individuals, as well as important differences in components of the immune system that are not measured in the data.” It’s like trying to find what you need in a haystack, but with more clear information on the path to finding it. ”
Despite benefiting from antiretroviral therapy, the HIV-positive group had clear differences in vaccine-induced responses compared to the control group, and the machine learning model was able to classify those differences with nearly 100% accuracy, although there were two people who were unable to distinguish them from the control group.
“No matter how we shuffled the data, no matter what biomarkers we used, machine learning algorithms could not distinguish between a small subset of people who were HIV-positive and people who were HIV-negative,” Kolosek says. “In these people, the immune response induced by the vaccine was indistinguishable from the HIV-negative group. This suggests that their immune function was effectively restored, at least with regard to the vaccine response.”
Conversely, there was one person in the healthy control group whose markers appeared indistinguishable from those infected with HIV. This may indicate an underlying immune problem that has not yet been identified clinically.
Supported by the National Research Council of Canada (NRC), the Fields Collaborative Center for Mathematical Sciences, the National Science, Engineering and Research Council of Canada, and Artificial Intelligence for Public Health (AI4PH), the research is published today as a preprint in the journal Patterns and will be printed as a cover story on March 13. Korosek worked with collaborators including Heffernan and Mohammad Sajjad Ghaami, a senior researcher at NRC Digital. Associate Professor Jessica Conway of the Center for Technology Research, Penn State University, and researchers at the University of Toronto and St. Michael’s Hospital.
“This study brings us closer to understanding the diversity of immunity in people living with HIV, how their responses compare with age-matched controls, how well antibodies are maintained over time, and why some people show markedly different patterns,” Kolosek says.
/Open to the public. This material from the original organization/author may be of a contemporary nature and has been edited for clarity, style, and length. Mirage.News does not take any institutional position or position, and all views, positions, and conclusions expressed herein are those of the authors alone. Read the full text here.
