Single-cell technologies generate high-dimensional data that reveal immune heterogeneity, yet linking cellular alterations to disease outcomes remains challenging due to data complexity and limited model interpretability. To address this, we developed computational approaches, including an explainable AI framework to identify cell-specific drivers of clinical outcomes. We also designed a single-cell longitudinal simulator with methods to detect temporal dynamics and disease interaction effects. Applying these approaches to autoimmune disease, with rheumatoid arthritis as an example, we uncovered circulating immune phenotypes and predictive signatures of disease onset, highlighting the potential of interpretable modelling for biomarker discovery and disease prevention.