Machine-learning immune-system analysis study may hold clues to personalized medicine

Machine-learning immune-system analysis study may hold clues to personalized medicine

York U led study found clear vaccine-initiated immune response biomarkers between HIV positive and HIV negative groups, but outliers underscore varied, intricate nature of the immune system

TORONTO, March. 4, 2026 – How people with compromised immune systems respond to vaccines is an important area of immunological research. A new study led by York University found that not only could machine-learning models accurately pinpoint differences in healthy controls and those living with HIV, but also found outliers in both groups that provide fascinating glimpses into the complex nature of the immune system and what personalized medicine could look like in the future, accounting for variables such as age, comorbidities and genetics.

“This study constitutes an important step forward in the potential for personal vaccination intervention strategies,” says lead author Chapin Korosec, who worked on this paper as a postdoctoral fellow at York University under the supervision of Faculty of Science Professor Jane Heffernan, whose research focuses on infectious disease modelling. “By learning the structure of immune variability at scale, we move toward a data-driven foundation for personalized vaccination and therapeutic design.”

Korosec, now an adjunct professor with the University of Guelph, used a dataset of people with and without HIV who had received up to five doses of COVID-19 vaccine over the course of 100 weeks. All the individuals living with HIV were from the Greater Toronto Area whose illness was being controlled with antiretroviral therapy. The researchers used a type of machine-learning method called random forest to analyze 64 immune biomarkers elicited through a response to the COVID-19 vaccine, and then created a group of ‘virtual patients’ to further model immune responses.

“While we were working with a rich dataset well suited for statistical testing, longitudinal mathematical models still face identifiability limits when the data cannot uniquely resolve immune dynamics. We therefore turned to machine learning to identify the core differences between groups, and then leveraged that learned structure to generate virtual patients that capture how immune patterns differ between groups.”

They were able to show that saliva-based antibodies, particularly a type of antibody in the saliva called IgA, coupled with white blood cells, which have long been known to be associated with HIV status, create the signature difference between the two groups. Korosec says this is significant because there is a lot of research showing altered mucosal immunity for those living with HIV and how it is influenced in the short and long term.

Heffernan notes 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 variability.

“The immune response is very, very complicated.” explains Heffernan. “Sometimes something can act as an inhibitor of an arm of the immune response, but in other times it might be an activator. There is also a lot of individual variability among people with similar immune system status. Using machine learning, mechanistic modelling, and ‘virtual patients’ we can try to uncover important differences in the subgroups and between individuals – even of immune system components that are not measured in the data. Kind of like trying to find the needed in a haystack, but with a clearer path to finding it.”

The HIV positive group, despite having the benefits of antiretroviral therapy, had clear differences in their vaccine-elicited responses compared to the control group and the machine-learning model was able to classify those differences with nearly 100 per cent accuracy, but there were two individuals who they could not differentiate from the control group.

“No matter how we shuffled the data or which biomarkers we used, the machine-learning algorithm could not distinguish a small subset of HIV-positive individuals from those who were HIV-negative,” says Korosec. “In those individuals, the vaccine-induced immune responses were indistinguishable from the HIV-negative group. That suggests that, at least in terms of vaccination response, their immune function was effectively restored.”

Conversely, there was one individual in the healthy control group whose markers looked indistinguishable from someone living with HIV, which may suggest underlying immune issues that may not yet have been clinically identified.

Supported by the National Research Council of Canada (NRC)-Fields Mathematical Sciences Collaboration Centre, the National Sciences and Engineering and Research Council of Canada and Artificial Intelligence for Public Health (AI4PH), the study was published today as a pre-print in the Journal Patterns and will appear in print as the cover article on March 13. Korosec worked with collaborators, including Heffernan,  Senior Research Officer Mohammad Sajjad Ghaemi from the NRC Digital Technologies Research Centre, Associate Professor Jessica Conway from Pennsylvania State University and researchers from the University of Toronto and St. Michael’s Hospital.

“This study moves us closer to understanding immune diversity in people living with HIV; how their responses compare to age-matched controls, how well antibodies are maintained over time, and why some individuals show strikingly different patterns,” says Korosec.

 

-30-

 

York University is a modern, multi-campus, urban university located in Toronto, Ontario. Backed by a diverse group of students, faculty, staff, alumni and partners, we bring a uniquely global perspective to help solve societal challenges, drive positive change, and prepare our students for meaningful life and career paths. York's Glendon Campus is home to Southern Ontario's Centre of Excellence for French Language and Bilingual Postsecondary Education. York’s campus in Costa Rica offers students exceptional transnational learning opportunities and innovative programs. Together, we can make things right for our communities, our planet, and our future.

END