Generating Privacy-Preserving Synthetic Immunological Data Using Variational Autoencoders: Insights from Canadian HIV and SARS-CoV-2 Vaccination Studies

The interplay of humoral and cellular immune responses to vaccination is a dynamic and heterogeneous process, particularly in populations with preexisting health conditions. Understanding these dynamics requires large datasets, but the sensitive nature of clinical data poses challenges to sharing and analysis. In this study, we leverage Variational Autoencoders (VAEs) to generate physiologically accurate synthetic datasets from Canadian clinical data tracking immunological trends in HIV-positive individuals on cART and an age-matched HIV-negative control group. Participants underwent up to five SARS-CoV-2 vaccinations over 500 days, with longitudinal blood draws measuring T cell and plasma B cell activity, antibody levels, virus neutralization, and ACE2 displacement. The proposed VAE framework preserves the statistical properties of the original data while ensuring patient privacy, enabling broader data sharing and hypothesis testing. By accurately modeling complex, time-dependent interdependencies in the immune system, our synthetic datasets facilitate research into immune responses, vaccine efficacy, and health disparities. This approach demonstrates the potential of synthetic data to address privacy concerns and inequities in electronic health data, providing a scalable solution for advancing personalized immunological research.