An open benchmark suite, harmonized kidney cell atlas, and community portal for AI-driven drug target discovery — developed at Johns Hopkins University to address the most under-treated global health crisis.
Acute kidney injury affects over 40 million people worldwide, causing over 6 million deaths annually, yet it remains the most under-addressed global health crisis. In hospitals, AKI drives morbidity and cost; in low- and middle-income countries, it is community-acquired, affecting younger populations. Despite two decades and $2B+ in failed trials, there are zero FDA-approved therapies to prevent AKI-to-CKD progression.
Even small AKI episodes drive CKD through repeated micro-injury, making AKI the principal engine fueling the global rise of kidney disease. Over 850 million people live with kidney disease; each progression to dialysis costs $90K+/year. The window between acute injury and chronic progression remains poorly understood and entirely untargeted by existing therapies.
Powered by multi-modal AI integrating single-cell and spatial tissue biology, human genetics, clinical data, and drug chemistry — providing standardized evaluation tasks, confidence-scored predictions, and a self-service portal.
Curated integration of publicly available single-nucleus transcriptomes from KPMP, NEPTUNE, ERCB, and GTEx, with spatial tissue maps and plasma proteomics — standardized to shared ontologies (HGNC, MONDO, CL).
snRNA-seq · Proteomics · GWASEight-dimension Target Biology Profile (TBP) engine: genetic evidence, disease validation, druggability, safety, homeostatic dispensability, development feasibility, competitive differentiation, and biomarker potential.
MR · eQTL · Open Targets · ChEMBLInteractive web portal where researchers explore the atlas, submit model predictions, benchmark against ground truth labels, and compare approaches on standardized tasks.
Open Access · Leaderboard · APIMulti-modal AI with causal inference (Mendelian randomization, gene regulatory networks), Bayesian uncertainty quantification, and privacy-preserving federated learning — every prediction carries a confidence score.
Causal Inference · Bayesian UQ · Federated LearningExplore representative visualizations from our pilot analyses — built on genomics, transcriptomics, proteomics, phenomics datasets and published literature.
UMAP projection of major renal cell types from snRNA-seq data (representative visualization)
The Target Biology Profile (TBP) engine scores each candidate across 8 orthogonal dimensions. Below are illustrative profiles for select targets using publicly available evidence.
Over two decades and $2B+ invested, more than 20 drug programs targeting AKI have failed in clinical trials. Renal PRISM systematically analyzes why — and what the field can learn.
Phase 2 & 3 interventional trials currently recruiting or active — a landscape Renal PRISM aims to inform
Pilot analyses integrating publicly available single-cell transcriptomics, proteomics, and genetic causal inference are generating new insights into AKI-to-CKD biology — and revealing patterns in why prior therapeutic approaches have not succeeded.
Computational analysis of kidney injury transcriptomes suggests the existence of distinct molecular subtypes within the AKI-to-CKD transition, potentially requiring different therapeutic approaches.
Pilot scoring across 8 dimensions — genetic evidence, druggability, safety, disease validation, and more — has prioritized a portfolio of candidates spanning novel mechanisms, known pathways, and biomarker opportunities not previously linked to AKI progression.
Multi-cohort plasma proteomics analysis has identified circulating proteins that distinguish AKI patients who will progress to CKD from those who recover — enabling earlier intervention and precision trial enrichment strategies.
Targeting ~2,500 researchers, 30+ institutions, and 10+ industry labs — with all datasets, models, and benchmarks freely available to the global research community.
Harmonized atlas and benchmark datasets released under permissive licenses. FAIR-compliant metadata with full provenance tracking.
All AI models, training code, and evaluation scripts published on GitHub and Hugging Face. Reproducible end-to-end pipelines.
Interactive portal for dataset exploration, model benchmarking, and collaborative target validation. No gatekeeping — science should be open.