AI-driven big data analysis and predictive modeling of infectious disease immunity: from correlates to causal,...

The ability to reliably predict protective immunity against infectious diseases remains a central challenge in immunology, vaccine development, and public health. Despite major advances in immunological measurement, immunity prediction has historically relied on simplified correlates most notably antibody titers that often fail to generalize across pathogens, populations, and time. The emergence of large-scale, high-dimensional immune datasets spanning multi-omics, longitudinal cohort studies, electronic health records, and pathogen genomics has created new opportunities for artificial intelligence (AI)-driven modeling. However, early applications reveal that predictive accuracy alone is insufficient and is frequently undermined by spurious correlations, limited generalizability, and a lack of biological interpretability. This review synthesizes current advances and limitations in AI-based immunity prediction, arguing that meaningful progress requires embedding machine learning within immune biology, causal reasoning, and rigorous validation frameworks. We examine the biological foundations of immunity relevant to modeling including innate immune programming, trained immunity, adaptive immune dynamics, and immune memory and map these processes to appropriate data modalities and model classes. We critically evaluate predictive tasks across individual and population scales, from early infection outcomes and vaccine responsiveness to immunity waning and herd-level protection. Emphasis is placed on validation beyond internal cross-validation, on interpretability grounded in immune mechanisms, and on robustness under demographic diversity and pathogen evolution. We conclude that immunity is best understood and predicted as a dynamic, probabilistic, and multiscale system rather than as a static trait. AI has substantial potential to improve infectious disease preparedness and clinical decision-making.