Abstract: FR-OR014
NephroAIX: Large Language Model (LLM)-Guided Knowledge Graph for Explainable Target Discovery in Human Kidney Diseases Using Single-Cell Data
Session Information
- Artificial Intelligence and Data Science Transforming Kidney Care: From Algorithms to Action
November 07, 2025 | Location: Room 361A, Convention Center
Abstract Time: 05:10 PM - 05:20 PM
Category: Artificial Intelligence, Digital Health, and Data Science
- 300 Artificial Intelligence, Digital Health, and Data Science
Authors
- Hayat, Sikander, Universitatsklinikum Aachen, Aachen, NRW, Germany
- Joppich, Markus, Ludwig-Maximilians-Universitat Munchen, Munich, BY, Germany
- Kramann, Rafael, Universitatsklinikum Aachen, Aachen, NRW, Germany
Background
Kidney diseases represent a significant global health burden and often progress to end-stage renal disease with limited treatment options. Drug discovery is complex, time consuming and error prone. Thus advanced explainable artificial intelligence (XAI) methods are needed to accelerate precision target discovery.
Methods
We created NephroAIX knowledge graph (KG) containing 2.1 million associations from drug, gene, disease, pathways and metabolites databases. Kidney Precision Medicine Project (KPMP) single-cell transcriptomics data from ~300K and approx. 100 human samples covering AKI, covid-AKI, CKD and diabetic kidney disease was projected onto the NephroAIX KG (170217 nodes, 2130677 edges). NeproAIX identifies condition-specific subnetworks derived through up-/down- stream ligand-receptor interactions, in order to identify potential drugs targeting these networks. NephroAIX uses Llama-3, a large language model (LLM) for systematic target prioritization.
Results
Overall NephroAIX identified 1937 disease-associated sub-networks. Using network-based community detection algorithms, NephroAIX identified FGFR-NPR1, and JAML-CXADR as the top two interactions in immune cells and podocytes in AKI. LUCITANIB, a multi-kinase inhibitor targeting growth factor receptors VEGFRs and FGFRs, and VESENCUMAB, a monoclonal antibody inhibiting VEGFR-2 were identified as top repurposing candidates to target FGFR-NPR1 axis in AKI. For CKD, we found ATN-161, a peptide antagonist of integrin α5β1, a receptor involved in cell-matrix adhesion, inflammation, and fibrosis, and VOLOCIXIMAB, a chimeric monoclonal antibody that targets integrin α5β1 as two drugs associated with immune-podocyte interactions. Focusing on the extracellular matrix, BIMOSIAMOSE was found as the top hit for modulating VCAN-SELL interaction by targeting Selectin L that mediates leukocyte tethering on vascular endothelium.
Conclusion
NephroAIX predicts novel targets and generates interpretable features such as genes, pathways, drugs, clinical trials and other disease-drug-target related information. Integrating AI with multi-modal single-cell omics holds immense potential to identify novel targets, elucidate disease pathways, and accelerate drug discovery. This synergy paves the way for precision nephrology, transforming how kidney diseases are understood and treated.
Funding
- Government Support – Non-U.S.