Multimodal predictions of end stage chronic kidney disease from asymptomatic individuals for discovery of genomic biomarkers

Abstract

Chronic kidney disease (CKD) is a complex condition where the kidneys are damaged and progressively lose their ability to filter blood, 10% of the world population have the disease that often goes undetected until it is too late for intervention. Using the UK Biobank (UKBB) we constructed a CKD cohort of patients (n=46,986) with genomic, clinical and demographic data available, a subset (n=2,151) having also whole body Magnetic Resonance Imaging (MRI) scans. We used this multimodal cohort to successfully predict, from initially healthy patients, their 5-year outcomes for End-Stage Renal Disease (ESRD, n=210, AUC=0.804 {+/-} 0.03 with 5 fold cross-validation) and the larger cohort for validation to predict time-to ESRD and perform Genome-wide association studies (GWAS). Extracting important clinical, phenotypic and genetic features from the models, we were able to stratify the cohorts based on a novel set of significant previously unreported SNPs related to mitochondria/cell death, kidney development and function. In particular, we show that the risk allele of SNP rs1383063 present in 30% of the population irrespective of ancestry and putatively regulating MAGI-1, a gene expressed in the podocyte slit diaphragm, is a strong predictor of ESRD and stratifies male populations of older age.