Waning immunity following Bordetella pertussis (whooping cough) vaccination remains a major obstacle to achieving long-term protection. This challenge is particularly pronounced with acellular pertussis (aP) vaccines, which, while safer and better tolerated than the earlier whole-cell pertussis (wP) formulations, induce shorter-lived immunity and weaker memory responses. Although booster immunizations are deployed to restore immune protection, the temporal dynamics and mechanisms that govern post-booster immune responses remain poorly understood. Here, we present a data-driven, multimodal computational framework that integrates longitudinal gene expression, immune cell populations, antibody titers, and cytokine concentrations to characterize post-booster immune responses. Leveraging baseline pre-booster measurements, we employed ensemble learning model to forecast time-evolved immune trajectories, achieving high predictive performance across an independent validation cohort. Our model identifies molecular and cellular signatures predictive of durable immunity, including distinct transcriptional modules and immune subsets associated with robust recall responses. This integrative approach advances beyond traditional single-endpoint analyses by capturing the dynamic interplay among immune system components over time. These insights not only illuminate mechanisms of long-lasting protection but also provide a framework for rational vaccine design. Our work demonstrates the power of machine learning in immunological modeling, establishing a benchmark for applying AI to vaccine response prediction and accelerating the development of next-generation pertussis vaccines with sustained efficacy.
