# Diabetic kidney disease: integrating multi-omics insights, artificial intelligence, and novel therapeutics for precision medicine

## TL;DR

This paper reviews how combining multi-omics data, AI, and new therapies can improve early detection and personalized treatment of diabetic kidney disease.

## Contribution

The paper introduces an AI-integrated framework that combines omics, imaging, and clinical data for precision medicine in diabetic kidney disease.

## Key findings

- Single-cell RNA sequencing and spatial transcriptomics reveal cellular heterogeneity and immune dysregulation in DKD progression.
- Metabolomics identifies early biomarkers linked to mitochondrial dysfunction and oxidative stress in DKD.
- AI models like DeepDKD and KidneyIntelX enable non-invasive screening and refined risk stratification for DKD.

## Abstract

Diabetic kidney disease (DKD) remains a leading cause of global morbidity and mortality. While current therapies like sodium-glucose cotransporter 2 (SGLT2) inhibitors and glucagon-like peptide-1 receptor agonists (GLP-1RAs) have improved outcomes, significant challenges persist in early detection and halting progression. This review synthesizes recent transformative advances in DKD research. We highlight how single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics have unveiled unprecedented cellular heterogeneity, delineated pathogenic trajectories like maladaptive tubular cell states, and established immune dysregulation as central to disease progression. Concurrently, metabolomics provides a window into early metabolic disturbances, identifying novel biomarkers that reflect mitochondrial dysfunction and oxidative stress. Furthermore, artificial intelligence (AI) is revolutionizing clinical practice, with deep learning models like DeepDKD demonstrating high accuracy in non-invasive screening using retinal images and enabling refined risk stratification. These multi-omics insights are paralleled by the development of novel therapeutic agents targeting inflammation, fibrosis, and metabolic pathways beyond traditional targets. The integration of high-resolution molecular profiling, AI-driven analytics, and mechanism-based therapeutics is paving the way for a new era of precision nephrology, offering hope for earlier intervention and personalized management strategies for DKD.

AI-integrated framework for clinical applications in DKD. The figure presents a three-column workflow: Data sources (left, yellow) include omics (scRNA-seq, spatial-omics, metabolomics), imaging (retina, renal biopsy), and clinical/EHR data (labs, phenotypes, KDIGO staging). The AI integration engine (middle, pink) features feature engineering, ML/DL algorithms (XGBoost, random forest, DNN), and explainability tools (SHAP, LIME, attention maps). Clinical outputs (right, blue) progress from early screening (retina-based DeepDKD) to risk stratification (beyond KDIGO using KidneyIntelX) and targeted therapies (SGLT2i, GLP-1RAs, ns-MRAs). The white-background diagram uses color-coded sections and directional arrows to show data flow from inputs through AI processing to clinical implementation.Flowchart illustrating pathways from data source to clinical output. Data sources include omics, imaging, and clinical/EHR data leading to an AI integration engine. The engine processes data through feature engineering, machine learning/deep learning algorithms, and an explainability module. Clinical outputs encompass early screening, risk stratification, and targeted therapy applications. The process is divided into discovery, validation and interpretation, and clinical implementation stages.

AI-integrated framework for clinical applications in DKD. The figure presents a three-column workflow: Data sources (left, yellow) include omics (scRNA-seq, spatial-omics, metabolomics), imaging (retina, renal biopsy), and clinical/EHR data (labs, phenotypes, KDIGO staging). The AI integration engine (middle, pink) features feature engineering, ML/DL algorithms (XGBoost, random forest, DNN), and explainability tools (SHAP, LIME, attention maps). Clinical outputs (right, blue) progress from early screening (retina-based DeepDKD) to risk stratification (beyond KDIGO using KidneyIntelX) and targeted therapies (SGLT2i, GLP-1RAs, ns-MRAs). The white-background diagram uses color-coded sections and directional arrows to show data flow from inputs through AI processing to clinical implementation.