Proteomics Reveals Novel Biomarkers in the Blood for Sepsis Diagnosis

Abstract

Early sepsis diagnosis in critically ill patients remains a major unmet need. We employed a multiphase proteomic strategy integrating discovery by data-independent acquisition, targeted verification by parallel reaction monitoring, clinical confirmation with ELISA, and machine learning to identify robust blood biomarkers. In discovery and verification cohorts, three consistently downregulated proteins, α-2-HS-glycoprotein (AHSG), clusterin (CLU), and kallistatin (SERPINA4), were prioritized. A random forest model combining these markers yielded an apparent area under the curve (AUC) of 0.94 in the training cohort. Crucially, in a prospective evaluation of 143 patients with suspected sepsis, representing the true test of generalizability, the model achieved an AUC of 0.87, significantly outperforming PCT and CRP, and correctly classified 73.0% of confirmed sepsis cases as high-risk. This triple-biomarker panel enables rapid, accurate early detection of sepsis and offers a readily translatable diagnostic tool for clinical practice.