Multimodal predictions of end stage chronic kidney disease from asymptomatic individuals for discovery of genomic biomarkers
A genetic variant affecting kidney filtering cells predicts who might develop kidney failure five years early, potentially helping doctors identify at-risk people across all ancestry groups.
Using UK Biobank data from 46,986 CKD patients with genomic, clinical, and MRI data, IBM researchers developed a multimodal ML model achieving AUC 0.804 for 5-year ESRD prediction from initially healthy patients. GWAS identified a novel SNP rs1383063 in MAGI-1 (a podocyte diaphragm gene), present in 30% of the population regardless of ancestry, as a strong ESRD predictor with potential utility for population-level kidney risk stratification.
What the study was
- Study design
- Retrospective cohort study with GWAS; UK Biobank multimodal (genomic + clinical + MRI)
- Population
- UK Biobank CKD cohort
- Sample size
- 46986
- Category
- Diagnostics
- Maturity
- Exploratory
- Journal
- BMC Nephrology
Why it surfaced
Large-scale multimodal ML with genomic biomarker discovery from UKBB; novel SNP with population-level frequency is clinically significant. Retrospective design and lack of prospective validation limit score. Topic is CKD rather than core cancer/hematology watchlist but relevant to AI/ML and genomics topics.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.