Targeted quantitative lipidomic uncovers lipid biomarkers for predicting the presence of compensated cirrhosis and discriminating decompensated cirrhosis from compensated cirrhosis

Abstract

Objectives: This study aimed to characterize serum lipid metabolism and identify potential biomarkers for compensated cirrhosis (CC) predicting and decompensated cirrhosis (DC) discrimination using targeted quantitative lipidomics and machine learning approaches.

Methods: Serum samples from a cohort of 120 participants was analyzed, including 90 cirrhosis patients (45 CC patients and 45 DC patients) and 30 healthy individuals. Lipid metabolic profiling was performed using targeted LC-MS/MS. Two machine learning methods, least absolute shrinkage and selection operator (LASSO), and random forest (RF) were applied to screen for candidate metabolite biomarkers.

Results: The metabolic profiling analysis showed a significant disruption in patients with CC and DC. Compared to the CC group, the DC group exhibited a significant upregulation in the abundance of glycochenodeoxycholic acid (GCDCA), glyco-ursodeoxycholic acid (GUDCA), glycocholic acid (GCA), phosphatidylethanolamine (PE), N-acyl-lyso-phosphatidylethanolamine (LNAPE), and triglycerides (TG), and a significant downregulation in the abundance of ceramides (Cer) and lysophosphatidylcholines (LPC). Machine learning identified 11 lipid metabolites (abbreviated as BMP11) as potential CC biomarkers with excellent prediction performance, with an AUC of 0.944, accuracy of 94.7 %, precision of 95.6 %, and recall of 95.6 %. For DC discrimination, eight lipids (abbreviated as BMP8) were identified, demonstrating strong efficacy, with an AUC of 0.968, accuracy of 92.2 %, precision of 88.0 %, and recall of 97.8 %.

Conclusions: This study unveiled distinct lipidomic profiles in CC and DC patients and established robust lipid-based models for CC predicting and DC discrimination.

Keywords: biomarkers; cirrhosis; decompensated; lipids; machine learning.