High-dimensional spatial proteomics and novel machine learning pipeline identifies disease specific renal damage...

Area of Science:

• Nephrology
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Lupus nephritis (LuN) and renal allograft rejection (RAR) cause kidney failure through inflammation and fibrosis.
• Quantitative assessment of spatial injury patterns in kidney diseases is crucial for understanding disease progression and developing targeted therapies.

Purpose of the Study:

• To develop and validate a computational pipeline, Renal Damage diagnosis (RDDx), for quantitative spatial analysis of kidney biopsies.
• To differentiate and quantify injury mechanisms in LuN and RAR using high-dimensional spatial proteomics data.

Main Methods:

• Collected spatial proteomics data from 23 LuN, 33 RAR, and 8 kidney control (KC) biopsies.
• Developed three neural networks for Renal Damage diagnosis (RDDx) to segment and classify kidney structures (tubules, capillaries, glomeruli) in whole-slide images.
• Validated RDDx accuracy, generalizability, and correlation with histopathological scores.

Main Results:

• RDDx accurately segmented kidney structures and identified tubular and vascular inflammation.
• Both LuN and RAR showed reduced tubular/capillary areas and expanded interstitial space.
• LuN exhibited patchy stressed tubules; RAR showed diffuse injury with specific patterns for T cell-mediated rejection (TCMR) and antibody-mediated rejection (ABMR).
• RDDx metrics correlated with histopathological scores, demonstrating clinical relevance.

Conclusions:

• RDDx effectively identifies and quantifies disease-specific damage mechanisms in LuN and RAR.
• Stressed tubules and proliferating capillaries are sensitive markers for LuN and ABMR, respectively.
• RDDx facilitates mechanistic studies and therapeutic target discovery in kidney diseases.