Characterization and classification of chronic kidney disease by spatial MIST and deep learning algorithm

Chronic kidney disease (CKD) is characterized by disruption of the native kidney architecture at the cellular and molecular levels, leading to eventual kidney fibrosis. To better resolve the spatial complexity of fibrotic remodeling, we applied spatial multiplexed immunostaining with signal tagging (Spatial MIST), a high-dimensional proteomic platform capable of quantifying protein expression at single-cell resolution across intact human kidney tissue specimens. Using kidney biopsies from control/low-grade and high-grade fibrosis, we profiled 22 protein markers to assess structural alterations, cell-type distribution, and spatial relationships across glomerular and interstitial compartments. Spatial proximity analysis revealed fibrosis-associated reorganization of endothelial and epithelial markers, including increased separation between CD31 and β-catenin and altered clustering of podocyte and immune markers. Integration with unsupervised uniform manifold approximation and projection (UMAP) clustering distinguished discrete cell populations, whereas correlation analysis with kidney function metrics revealed that vimentin and alpha smooth muscle actin (α-SMA) positively correlated with fibrosis severity, whereas Wilms tumor 1 (WT1) expression was inversely correlated with declining kidney function. A graph neural network (GNN) classifier trained on spatial proteomic features further identified megalin, WT1, and vimentin as a top predictor of fibrosis grade. Together, these findings demonstrate the utility of Spatial MIST for capturing the molecular heterogeneity of CKD and uncovering spatial signatures of disease progression. This integrative approach provides a foundation for biomarker discovery and spatially informed classification of kidney pathology.<b>NEW & NOTEWORTHY</b> In this study, we offer a novel spatial analysis of markers relevant to CKD, which may provide useful insights into disease progression. By using this spatial proximity data, we created a GNN model that is capable of classifying disease severity and identifying markers that are most important for its classification. This integrative approach offers a foundation for future studies aimed at developing clinically actionable tools for CKD diagnosis and prognosis.