대한신장학회

논문분류	춘계학술대회 초록집
제목	Machine Learning Based Urinary Metabolome Profiling to Differentiate Primary Glomerular Diseases
저자	Seung Hyun Han
출판정보	2026; 2026(1):
키워드	glomerular disease, metoblomic profiling, machine learning, Biomarker panel
초록	Objectives: Glomerular diseases (GDs) comprise a biologically heterogeneous group of kidney disorders that frequently present with overlapping clinical manifestations, thereby limiting non-invasive subtype differentiation and reinforcing reliance on kidney biopsy. Although metabolomics provides an integrated view of disease-associated biological processes, its utility for subtype differentiation and molecular stratification across GDs remains insufficiently explored. Methods: Urinary metabolomic profiling was conducted in biopsy-confirmed GDs (IgA nephropathy [IgAN], focal segmental glomerulosclerosis [FSGS], minimal change disease [MCD], membranous nephropathy [MN]) from a nationwide multicenter cohort across six tertiary hospitals in Korea, including a discovery set (n = 1,060) and an independent validation set (n = 550). Metabolites were quantified using an IVDr-based ¹H NMR platform. Machine learning–based feature selection identified an eight-metabolite panel for disease discrimination, which was externally validated in the independent cohort. Results: Urinary metabolomic profiles differed across glomerular disease subtypes, providing additional metabolic characterization beyond conventional clinical factors. Machine learning–based feature selection yielded eight candidate metabolites for disease discrimination. The eight-metabolite panel achieved excellent discrimination between controls and GDs (area under the receiver operating characteristic curve [AUC] 0.983) and demonstrated strong subtype classification for IgAN (AUC 0.829), MN (0.842), and MCD (0.823), whereas performance was comparatively limited for FSGS (0.681). Overall multiclass performance was robust (macro-average AUC 0.831). Performance also remained stable when diabetic nephropathy was included as a disease control (macro-average AUC 0.850). Consistent performance was observed in the independent external validation cohort Conclusion: Urinary metabolomic profiling identifies subtype-associated metabolic signatures across glomerular diseases. These findings suggest the potential of machine learning–derived metabolite panels as a complementary non-invasive approach to disease stratification.
원문(PDF)	PDF 원문보기

간행물 검색

대한신장학회 회원관리 규정