MACHINE learning is helping scientists uncover how COVID-19 vaccines trigger different immune responses in people living with HIV, offering new insights that could support more personalised vaccination strategies.
Vaccines work by training the immune system to recognise and fight infections. However, immune responses can vary widely between individuals, particularly in people with underlying conditions such as HIV, which affects immune function. Understanding these differences is important for ensuring vaccines provide strong and lasting protection in vulnerable groups.
In the new study, researchers used machine learning models to analyse complex immune data from older adults receiving up to five SARS-CoV-2 vaccinations. Participants included people living with HIV who were taking antiretroviral therapy as well as HIV-negative individuals of similar age.
Machine Learning Identifies Distinct Immune Signatures
Using a random forest machine learning approach, the research team examined multiple immune markers, including cytokines, antibodies, and indicators of cellular immune activation. The analysis revealed distinct immune signatures separating vaccinated people living with HIV from HIV-negative participants.
One key finding was that combinations of cytokines and saliva-based antibodies were particularly effective at distinguishing immune responses in people with HIV. These signals appeared to reflect differences in T cell activation and mucosal immunity, two important components of the body’s defence against viral infection.
Interestingly, traditional blood antibody measurements alone were less informative in identifying these differences. This suggests that saliva-based immune markers could provide additional insight into vaccine responses in immunocompromised populations.
Evidence of Immune Recovery in Some Patients
The machine learning analysis also revealed encouraging results. A subset of people living with HIV showed immune response patterns that closely resembled those of HIV-negative participants after vaccination. Researchers interpret this as evidence that effective antiretroviral therapy may restore aspects of immune function sufficiently to produce near-normal vaccine responses in some individuals.
Visualisation techniques, including dimensionality-reduction mapping, further highlighted clusters of immune features associated with different patterns of humoral and cellular immunity.
Synthetic Data and Future Precision Vaccination
To expand the study’s potential applications, researchers generated privacy-preserving “virtual patients” that reproduced the patterns seen in the original immune data. Machine learning models trained on these synthetic datasets were able to accurately classify immune responses in real patients, suggesting a promising approach for research while protecting sensitive health information.
Overall, the findings demonstrate how machine learning can help decode complex vaccine immunology. By identifying key immune features that shape vaccine responses, the approach could support more targeted monitoring and precision vaccination strategies for people living with HIV and other immunologically vulnerable populations.
Reference
Korosec CS et al. Modeling of longitudinal immune profiles reveals distinct immunogenic signatures following five COVID-19 vaccinations among people living with HIV. Patterns. 2026; DOI 10.1016/j.patter.2025.101474.
Featured image: Yura on Adobe Stock
