Transcriptomic and computational approaches to characterize and predict immune responses to vaccination: a multi-study perspective

Vaccines are highly effective, but they remain difficult to optimize and to personalize because vaccine responses are heterogeneous: individuals differ in reactogenicity, in response quality, and especially for therapeutic vaccines in clinical benefit. Systems vaccinology offers a way to manage this complexity, but only if high-dimensional measurements can be translated into biological understanding and clinically useful prediction. This thesis develops and applies integrative transcriptomic and computational approaches across three vaccine contexts to address fundamental questions in vaccine immunology: (1) What cellular and molecular mechanisms drive vaccine-associated adverse events? (2) Can baseline immune profiles predict therapeutic vaccine responses in chronic infections? (3) How do innate immune activation and adaptive B-cell responses coordinate to generate vaccine-induced protection? Methodologically, we introduce two frameworks: an interaction-adjusted model to separate true interaction signals from additive co-culture responses, and TENTACLES, a consensus machine learning pipeline to detect compact, reproducible gene signatures across heterogeneous cohorts. In the rVSVΔG-ZEBOV-GP Ebola vaccine, transcriptomics and in vitro infection models show tissue relevant tropism and reveal that arthritis-linked inflammatory programs emerge from direct monocyte–synoviocyte interactions rather than from either cell type alone. Interaction-specific programs integrate innate immune signaling with extracellular matrix remodeling and bone associated pathways, indicating that localized post-vaccination arthritis reflects tissue-specific immune–stromal crosstalk rather than generalized antiviral inflammation. In post kala-azar dermal leishmaniosis (PKDL), therapeutic vaccination with ChAd63-KH, clinical trials show robust immunogenicity but modest efficacy, motivating baseline stratification. Using TENTACLES, we identify a minimal baseline gene signature associated with subsequent clinical improvement across cohorts, supporting the concept that pre-treatment immune state can be an outcome determinant. In the iNTS-GMMA vaccine, integrating whole-blood transcriptomics with B-cell receptor features reconstructed from bulk RNA-seq captures coordinated kinetics: rapid, dose-dependent innate activation within 24 hours followed by day-7 B-cell programs, alongside evidence of vaccine associated clonal expansions and convergent (“public”) clonotypes shared across recipients. By integrating transcriptomic profiling with new computational methods across diverse vaccine platforms, this thesis shows that biological variability is not noise but an informative signal. Vaccine heterogeneity can be decomposed into interaction-driven tissue responses, baseline host state, and B-cell clonal dynamics, giving both biological insight and predictors relevant to clinical outcomes. We show that (1) adverse events reflect the combined effects of tissue tropism and immune–stromal communication; (2) baseline profiles can stratify patients more likely to benefit from therapeutic vaccination; and (3) transcriptomic and B-cell receptor data together helps track the immune response over time after vaccination and how the two arms work together. Overall, these approaches can be used in vaccine trials to interpret safety signals more clearly and support biomarker-based, personalized vaccination.