Modeling of longitudinal immune profiles reveals distinct immunogenic signatures following five COVID-19 vaccinations among people living with HIV

The immune response to vaccination is highly heterogeneous and arises from a dynamic interplay of immune components. Harnessing machine learning (ML) to learn immune interdependencies offers the potential not only to decode immune signatures linked to a specified comorbidity but also to reveal individualized patterns laying the groundwork for precision-guided vaccination and targeted clinical follow-up. We employ a random forest (RF) approach to classify informative differences in immunogenicity between older people living with HIV (PLWH) on antiretroviral therapy (ART) and an age-matched control group who received up to five SARS-CoV-2 vaccinations. RFs identify evidence for T helper 1 (Th1) imprinting and reveal novel distinguishing immune features, such as saliva-based antibody screening, as promising diagnostic tools (whereas serum IgG is not). Our modeling approach reveals a subset of PLWH whose immune signatures are indistinguishable from the HIV- control group, which we interpret as near-complete immune restoration from a longitudinal vaccine-elicited immunogenic perspective. To expand the utility of our findings, we generate privacy-preserving synthetic "virtual patients" that accurately approximate the original longitudinal immunologic data and show, via train-on-synthetic/test-on-real evaluation, that RF classifiers trained solely on virtual patients generalize to held-out real patients. Our results highlight the effectiveness in utilizing informative immune feature interdependencies for classification tasks and suggest broad impacts of ML applications for personalized vaccination strategies among high-risk populations.