黃佩瑜

Abstract:

Aberrant glycosylation is a hallmark of chronic inflammation and fibrosis. In this study, we introduce a label-free technique termed Wax Physisorption Kinetics-based Fourier Transform Infrared (WPK-FTIR) imaging to spatially resolve glycan remodeling on glomerular surfaces during the progression of glomerulonephritis. This method leverages n-alkane probes of varying chain lengths (e.g., n-C₂₂H₄₆ and n-C₂₈H₅₈) to assess glycan elongation based on their hydrophobic physisorption profiles. Quantitative analysis of the absorbance ratio A(n-C₂₂H₄₆)_residue / A(n-C₂₈H₅₈)_residue revealed a progressive increase across disease states: normal (0.46 ± 0.06) → inflamed (0.95 ± 0.08) → fibrotic (1.54 ± 0.17) glomeruli, reflecting stage-specific glycan alterations. To extract biologically relevant information from the hyperspectral dataset, we applied Principal Component Analysis (PCA) to reduce dimensionality while retaining critical spectral variance. PCA loading vectors were interpreted to identify biochemical signatures, including glycogen-related C–O–H, and C–O–C vibrations, amide I/ II/ III protein bands, and lipid/ phosphate moieties. K-means clustering was then used to define unsupervised spectral phenotypes, which served as pseudo-labels for training supervised classifiers. A multi-stage machine learning pipeline was established using both Random Forest (RF) and Support Vector Machine (SVM) models to classify inflammation stages. This hybrid approach improved both classification accuracy and biological interpretability. In conclusion, the integration of WPK-FTIR imaging with unsupervised-to-supervised learning establishes a robust platform for label-free detection and stratification of glycan-driven pathological changes in kidney disease, offering new potential for early diagnosis and monitoring of therapeutic response.

Keywords – WPK-FTIR, Machine Learning, Kidney Disease, Glycan

 Back