Goderich resident authors study looking at vaccine outcomes in immunocompromised populations

Results of a recent study published in Patterns, a Cell Press Journal, reveal a mechanistic framework for understanding vaccine outcomes in immunocompromised populations.

Led by Goderich resident and study lead author, Chapin Korosec, the study looked at how COVID-19 vaccination elicits distinct immune responses in people with HIV, versus those without HIV.

Vaccines are key to controlling the impact of viruses during a pandemic, yet immune responses vary between individuals.

People with immune compromised conditions such as Rheumatoid Arthritis or cancer have different responses not only to viruses but to treatments and vaccines.

Those living with HIV (PLWH) represent an important group of people whom vaccine efficacy and immune response remain under-characterised.

People living with HIV are at a much larger risk of severe outcomes to viral diseases, due to the immune compromising nature of HIV.

At the same time, in general, immune response is extremely complicated.

Korosec explains that our immune response to viruses and vaccines varies considerably across healthy individuals depending on how efficient our immune system is.

The body contains a lymphatic system. It runs parallel to our circulatory system (blood), but it is not pumped by the heart.

According to Korosec, the lymphatic system is the entire system of lymph nodes (there are 400-600 depending on the person), connected via the lymph (conceptually like veins but contain lymph fluid, not blood).

Within each of our lymph nodes is everything needed to mount an immune response to a virus, vaccine, cancer or bacterial infection.

In the lymph node, ‘helper T cells’ play the central role of coordinating the immune response. T cells turn the immune system on, tell all other cells through chemical signals about the threat, and ultimately turn off the response when the threat has been dealt with.

Helper T cells are the immune system’s control board managers, coordinating which responses to turn on, how strong they are, and when to shut them down.

“HIV is a virus that specifically targets and infects these Helper T cells, so you can imagine how important it is to get that infection under control very quickly, or in our case, when it is under control with a drug regimen called ART, how well do vaccines work?” said Korosec.

This study was conducted to build an immune signature of vaccine responses in people living with HIV, compared to an age-matched, non-HIV group of individuals.

“In this way, we can use a machine learning model to pick up and learn the varied responses among individuals,” says Korosec.

“It learns the patterns that distinguish the individuals.”

A total of 91 participants were recruited into the study – 23 HIV-negative individuals and 68 living with HIV. Study participants were given five vaccine doses.

Data for the study was collected through the clinical lab of Dr. Mario Ostrowski and the University of Toronto.

All real people were used in the study, which helped build the virtual patient cohort.

Notably, the study demonstrates that virtual patients can be used to train models that generalise to real individuals. This supports precision-guided vaccination strategies.

While HIV remains a global health burden, with approximately 40 million current infections and more than 40 million deaths to date, millions of new infections are expected by 2050 according to the Global Burden of Disease Study 2021.

According to the study, individuals with advanced HIV disease are susceptible to more severe COVID-19 outcomes.

While vaccines are the intervention to reduce severity and morbidity of infectious diseases, the immunogenic response to COVID-19 vaccines among those living with HIV depends on the vaccine type, and disease progression.

More specifically, this study illustrates how data-driven immunology can inform precision vaccination strategies and improve continuity of care for immunologically compromised people.

The work from the study is an interdisciplinary effort with faculty of York, UofT (all data came from the lab of Dr. Mario Ostrowski, Dept. of Medicine), U Penn and National Research Council Canada, with Korosec as lead.

The team applied machine learning approaches to identify immunogenic signatures distinguishing vaccine responses in people living with HIV.

Korosec stresses there are broader implications for data-driven immunology, including vaccine science and AI applications in health.

According to Korosec, machine learning (ML) holds significant promise to revolutionizing HIV diagnostics by uncovering complex immune patterns and facilitating precision health interventions.

Knowledge produced from theses studies could inform customized vaccination strategies and reduce the risk of severe outcomes for immunologically vulnerable people.

AI (artificial intelligence) and ML (machine learning) models are becoming increasingly good at picking up and learning intricate patterns, while the immune system response remains very complicated.

According to Korosec, blood-based and saliva-based antibodies, the many different types of antibodies, the white blood cells and their subtypes and molecules that regular them all have difficult to predict responses.

Using ML or AI can allow scientists to make steps towards individualized healthcare. Both ML and AI can take in a set of data and learn how these complex responses are connected.

“Individualized healthcare means that your therapeutic or vaccine regimen is tailored especially to you, so that you as an individual get optimal outcomes to whatever medical intervention you are undertaking,” explained Korosec.

“In our work, we focused on immune compromised group – people living with HIV.”

ML and AI are two slightly different concepts.

AI is the broad field of creating systems that perform tasks requiring human-like intelligence, while ML is a subset of AI, which enables systems to learn patterns from data and improve their performance.

ML models specialise in learning patterns from data.

“You can, with enough data, pick out really complex connections,” added Korosec.

“In our case, between the many varied components of the immune system response.”

Results of the study revealed the power of ML to both create dominant response signatures and still identify outliers, that there is no different pattern between people with HIV and those without HIV in terms of producing antibodies despite differences in immune components, and using virtual patients can help scientists track how immune responses evolve over time.

Korosec explains that by using computer-generated virtual patients, outcomes can be better predicted, and better treatments or vaccines can be designed through simulation.

“If the models can capture the key relationships between the real data, and create accurate virtual data, we can synthesize virtual data for hypothesis testing to aid in future clinical studies,” explained Korosec.

Now that this study is complete and results have been printed in this journal, what’s next?

Korosec explains that much like how MRI and CT scanners are built by specialists, for AI models to be used by medical doctors, so should every use-case of AI models.

For example, it took a lot of time and specialist research to develop the specific brain scan protocols, or every specific organ scan protocol, used by an MRI machine.

Similarly, Korosec sees AI as an important tool in healthcare. However, he explains that every specific use-case of this tool needs to be built and tested before medical professionals can safely use it in their practice.

“Our research, as well as that of others, have shown that AI can be utilised in diagnosing outcomes and aiding in the individualization of healthcare,” explained Korosec.

“Administratively, there are hurdles to overcome, but many companies and research groups are working on this.”

Korosec furthers that the specific contribution from this study was to show that ML can discern immune differences from vaccination for people living with HIV.

“There is no reason why one could not do the same for individuals who smoke, or have other health issues such as diabetes or cancer,” Korosec added.

“For me, I would like to make my next step forward towards clinical implementation with real use-case scenarios.”

Chapin S. Korosec lives in Goderich and is an Adjunct Professor in the Department of Mathematics and Statistics at the University of Guelph.

Korosec’s article was selected to grace the cover for the March 2026 issue of the journal Patterns (by Cell Press).