Outlying immune responses: machine learning reveals a subset of HIV+ and HIV− individuals with atypical vaccine-elicited immune signatures

Understanding how people living with HIV (PLWH) respond to repeated COVID-19 vaccinations is critical for advancing precision medicine in immunocompromised populations. In this study, we use random forest models to identify which immune responses most effectively differentiate vaccine outcomes between PLWH on antiretroviral therapy and an HIV-negative control group. Our data set contains an extensive range of immune features, including serum and saliva IgG and IgA responses, ELISpot IFNg and IL2 responses to SARS-CoV-2 spike peptides, ACE2 receptor displacement, and SARS-CoV-2 neutralization capacity; all tracked longitudinally up to 104 weeks in each individual following SARS-CoV-2 vaccine dose 1, up to dose 5. Our model achieves near-perfect accuracy and reveals that cytokine-producing T cells and saliva-based IgA responses are key features for classification, whereas serum IgG markers offer limited classification value. Through ablation sensitivity analysis, we are able to identify outlier HIV- and HIV+ individuals whose immunological profiles do not fit the learned ‘pattern’ identified by the RF algorithm; some HIV+ individuals on ART appear to have nearly complete immune recovery while some HIV- individuals have vaccine-elicited immune signatures that appear like that of a typical HIV+ individual, suggesting immune compromisation.

Outlying immune responses: machine learning reveals a subset of HIV+ and HIV− individuals with atypical vaccine-elicited immune signatures

Monday, July 14 at 6:00pm

Chapin Korosec

"Outlying immune responses: machine learning reveals a subset of HIV+ and HIV− individuals with atypical vaccine-elicited immune signatures"

#SMB2025 Follow