<p>Core machine learning candidate biomarkers.</p>

Cross-species single-cell RNA-seq data from rat ileal tissue (GSE244963) were used only as supportive cell-type context for candidate genes.</p> <p>Results</p><p>GSEA identified enrichment of NETosis, complement/coagulation, oxidative phosphorylation, and multiple inflammatory cell death-related programs in the SAP group. ssGSEA revealed reduced adaptive immune-associated signatures, with CD8 + T-cell and NK-cell signatures already reduced in MAP and CD4 + T-cell and Treg-associated signatures showing additional reduction in SAP relative to MAP.

Triple-algorithm machine learning identified three core candidate genes with discovery-cohort AUC values: MRPL51 (ML-A, SAP vs HC, AUC = 0.951), C1QA (ML-B, SAP vs MAP, AUC = 0.754), and DACT1 (ML-B, SAP vs MAP, AUC = 0.758). Repeated stratified 5-fold cross-validation supported internal stability of the single-gene estimates, but no independent external validation cohort was available. A neutrophil-associated NETosis gene triad (OLFM4, LTF, and CEACAM6) was additionally co-selected by LASSO and random forest in ML-B. Rat ileal scRNA-seq provided supportive, cross-species cell-type context rather than definitive validation.</p> <p>Conclusion</p><p>Integrative transcriptomic analysis of peripheral blood identifies candidate transcriptomic markers and suggests neutrophil-associated, metabolic, and adaptive immune programs associated with AP severity progression.

The stratified adaptive immune signature reduction pattern and the NETosis-associated gene triad may represent discovery-level transcriptomic features of progression from mild AP to the severe-spectrum AP group. Independent validation in prospective human cohorts is required.</p></div>