Machine learning for identifying liver and pancreas cancers through comprehensive serum glycopeptide spectra analysis: a case-control study.

Abstract

Liver and pancreatic cancers are difficult to detect early, leading to high mortality rates. Blood-based diagnostics present a viable alternative for earlier detection, potentially improving survival rates. The comprehensive serum glycopeptide spectra analysis (CSGSA) method combines enriched glycopeptides (EGPs) with conventional tumor markers through machine learning to accurately identify early stage cancers. Here, we analyzed nine tumor markers (CA19-9, AFP, PSA, CEA, CA125, CYFRA, CA15-3, SCC antigen, and NCC-ST439) in 119 patients with pancreatic cancer and 49 with hepatocellular carcinoma, alongside 590 healthy controls. We also analyzed EGPs using liquid chromatography-mass spectrometry. We found that α1-antitrypsin with a fully sialylated biantennary glycan at asparagine 271 and α2-macroglobulin with a fully sialylated biantennary glycan at asparagine 70 effectively distinguished liver and pancreatic cancers. The integration of these two glycopeptides, along with the nine tumor markers and 1688 EGPs using a machine learning model enhanced diagnostic accuracy, achieving a receiver operating characteristic-area under curve (ROC-AUC) score of 0.996. CSGSA has the potential to minimize the need for invasive diagnostic procedures and serves as a promising tool for widespread screening.