Effective early prediction of acute pancreatitis (AP) severity remains an unmet clinical need due to limited molecular characterization of systemic immune responses. We performed integrated single-cell RNA sequencing with T- and B-cell receptor profiling on peripheral blood mononuclear cells from AP patients (n = 7) at days 1, 3, and 7 after admission. Immune landscape analysis revealed marked inter-patient heterogeneity, with a distinct expansion of MZB1-expressing plasma cells that were strongly associated with complicated AP and recovery. Functional validation in an independent cohort (n = 14) confirmed disease-associated plasma cell markers, alongside altered serum immunoglobulin and cytokine profiles (n = 32). From these findings, we established a nine-gene B-cell-derived transcriptomic signature (S100A8, DUSP1, JUN, HBA2, FOS, CYBA, JUNB, S100A9, and WDR83OS) predictive of AP severity. This model demonstrated high discriminative performance in internal validation (n = 114; AUROC > 0.95, superior to standard clinical scoring systems), and sustained accuracy in external validation cohorts of AP (n = 87) and AP combined with non-AP sepsis (n = 174) for predicting persistent organ failure. Our study identifies a mechanistic and predictive role for MZB1⁺ plasma cells in AP pathogenesis, offering a novel immune-based stratification strategy with potential for precision clinical management.
Keywords: B cells; acute pancreatitis; machine learning; molecular biomarker; severity prediction.
© 2025 The Author(s). MedComm published by Sichuan International Medical Exchange & Promotion Association (SCIMEA) and John Wiley & Sons Australia, Ltd.