Abstract: FR-OR016
Large-Scale Systemic Proteome Profiling of More Than 5,000 Proteins to Differentiate Primary Glomerular Diseases
Session Information
- Artificial Intelligence and Data Science Transforming Kidney Care: From Algorithms to Action
November 07, 2025 | Location: Room 361A, Convention Center
Abstract Time: 05:30 PM - 05:40 PM
Category: Artificial Intelligence, Digital Health, and Data Science
- 300 Artificial Intelligence, Digital Health, and Data Science
Authors
- Oh, Jae-ik, Seoul National University College of Medicine, Jongno-gu, Seoul, Korea (the Republic of)
- Jeong, Kyeonghun, Seoul National University, Gwanak-gu, Seoul, Korea (the Republic of)
- Park, Sehoon, Seoul National University Hospital, Jongno-gu, Seoul, Korea (the Republic of)
- Cho, Jeongmin, Chung-Ang University Gwangmyeong Hospital, Gwangmyeong-si, Gyeonggi-do, Korea (the Republic of)
- Kim, Kwangsoo, Seoul National University Hospital, Jongno-gu, Seoul, Korea (the Republic of)
- Kim, Dong Ki, Seoul National University College of Medicine, Jongno-gu, Seoul, Korea (the Republic of)
Background
Primary glomerular diseases (GN) require kidney biopsy for definitive diagnosis. We investigated whether systemic plasma proteome profiling with machine learning could enable non-invasive classification of GN subtypes.
Methods
We analyzed 5,416 plasma proteins using Olink Explore HT from a discovery and validation cohort. The discovery cohort included 14 patients with focal segmental glomerulosclerosis (FSGS), 46 with IgA nephropathy (IgAN), 48 with minimal change disease (MCD), 15 with anti-PLA2R antibody-positive membranous nephropathy (MN), and 38 healthy individuals, while the validation cohort included 14 FSGS, 32 IgAN, 15 MCD, and 10 MN patients. To develop models to classify each GN subtype, we trained logistic regression models with elastic net regularization by the discovery set with or without baseline clinical features (eGFR and UPCR), and evaluated performances by the validation set. Gene set enrichment analysis (GSEA) was performed on ranked protein coefficient values to identify associated biological pathways.
Results
Plasma proteome variation showed strong disease-specific patterns independent of eGFR or proteinuria. The model incorporating the top 50 proteins per subtype achieved strong discriminatory power, yielding an overall ROC-AUC of 0.86 in the validation set. Subtype-specific performance was also strong from MCD, MN, and IgAN (Figure 1C). However, performance was relatively lower for FSGS (AUC = 0.69). Notably, including UPCR and eGFR as features in the model did not significantly improve its performance. GSEA revealed enrichment of hemostasis in MCD, humoral immune response in IgAN, lymphocyte activation in FSGS, and regulation of cell differentiation in MN.
Conclusion
Distinct plasma proteome signatures exist for each GN subtype, providing a promising tool for non-invasive diagnosis and disease classification.