Development and Application of Machine Learning models for Systems Vaccinology in biological and clinical settings

Understanding the determinants of vaccine-induced immunity remains a central challenge in immunology, particularly in the context of novel vaccine platforms, emerging pathogens, evolving microbial variants, and necessity of protecting vulnerable populations. Although antibody titers are reliable and widely accepted correlates of protection, especially in many viral infections, they only partially reflect the broader and more complex dynamics of vaccine-induced immune responses. Systems vaccinology—a field at the intersection of immunology, omics technologies, and computational modeling—offers a holistic framework to dissect vaccine-induced immunity by leveraging high-dimensional data and dynamic biological measurements. This thesis explores how machine learning (ML) and mathematical modeling can be used to extract meaningful signatures from complex immunological datasets and generate actionable insights for vaccine development and deployment. The work is structured into four chapters, each addressing a key methodological or conceptual challenge in systems vaccinology. Novel ensemble learning frameworks are developed and applied across diverse experimental settings to identify robust immunological signatures and improve the stability and generalizability of biomarker discovery (Chapter 1). These approaches are validated in both diagnostic (Crohn’s disease and diagnosis of infection) and prognostic (therapeutic vaccine trials) contexts, as well as in preclinical models of infectious disease. ML-based approaches are leveraged to predict vaccine responsiveness in clinically fragile populations, including people living with HIV, patients undergoing hemodialysis, and recipients of solid organ transplantation (Chapter 2). Through supervised and unsupervised models applied to clinical, serological, and transcriptomic data, these studies identify interpretable predictors of immune response and highlight clinical factors that may inform vaccine scheduling, follow-up, and therapeutic decision-making in vulnerable groups. Vaccine-induced memory B-cells are investigated using high-dimensional flow cytometry and unsupervised ML tools (Chapter 3). By reconstructing cellular trajectories, comparing homologous and heterologous booster regimens, and modeling the persistence of memory B cells in both healthy and fragile individuals, the studies provide a nuanced understanding of memory generation and persistence beyond serum antibody titers. A novel mathematical framework is developed to isolate interaction-specific transcriptional signals from bulk RNA-sequencing of co-cultured cell types (Chapter 4). This approach is applied to an in vitro model of immune–stromal crosstalk following infection with the recombinant Ebola vaccine vector rVSVΔG-ZEBOV-GP. By separating cell–cell interaction effects from additive signals, the method enables the mechanistic investigation of pathways potentially involved in tissue-specific adverse events, such as post-vaccination arthritis. Together, these studies demonstrate the potential of computational tools not only to predict vaccine responses, but also to highlight the underlying biological processes that shape immunity. By integrating ML, systems immunology, and mechanistic modeling, these approaches contribute to the development of data-driven strategies that might help in guiding vaccination policies for future possible epidemic or pandemic.