POS-421 KIDNEYNETWORK: A NEW METHOD TO PREDICT KIDNEY DISEASE GENES USING KIDNEY DERIVED GENE EXPRESSION DATA IDENTIFIES A NEW CANDIDATE GENE FOR MILD ADPKD / PCLD

Abstract

Genetic testing of patients with suspected hereditary kidney disease using exome-based panel sequencing can reveal pathogenic variants in kidney-related genes. However, we are still faced with unsolved cases. Potentially harmful variants can reside in other genes that are either not annotated for kidney disease or in genes of unknown function. This makes it difficult to prioritize and interpret the relevance of variants in these genes for kidney diseases. As such, there is a clear need for methods that predict the phenotypic consequences of gene expression in a way that is as unbiased as possible. We have recently done this using public RNA sequencing (RNA-seq) data and shown that our method substantially improves the diagnostic yield of clinical exome sequencing in rare disease (Deelen et al., 2019). We have now extended upon this work by developing a kidney-specific gene network -KidneyNetwork- similar to www.genenetwork.nl using a combined approach of non-specific human RNA-seq data and human kidney-derived RNA-seq data. We calculated gene co-expression in 878 publicly available kidney RNA-seq samples and combined this with the calculated co-expression in the general human RNA-seq dataset. These expression patterns were used to predict genes involved in kidney related phenotypes as established in the HPO database. We used ‘leave-one-out’ cross validations to determine the prediction accuracy (AUC) of our predictions. The KidneyNetwork was applied to prioritize variants in 13 undiagnosed kidney patients for whom WES data were available. In KidneyNetwork, we observe a significantly improved prediction accuracy of kidney-related pathways as compared to GeneNetwork. Moreover, the total number of significantly predicted pathways has increased (Figure 1). For each HPO term, all genes are assigned a z-score representing their potential involvement in the respective pathway. Genes with a known association to that pathway are expected to have high scores and genes unknown for their involvement with high scores are potential new candidate disease genes. Based on the HPO terms “Polycystic kidney dysplasia” and “Hepatic cysts”, combined with information on potentially harmful variants in one of the undiagnosed patients with mild ADPKD/PCLD, we propose “gene X” (proper name will be revealed at the conference) as a new candidate gene. Gene X bears resemblance to other genes implicated in this phenotype. Two patients included in the Genomics England dataset with a similar phenotype have suspected deleterious variants in gene X. The improvement of KidneyNetwork relative to www.genenetwork.nl, has enabled us to more accurately predict candidate genes involved in renal disease. By analysing 13 undiagnosed patients, for which we found gene X (name will be revealed at the conference) to be a highly promising candidate gene, we have shown the added benefit of KidneyNetwork for the research community and the clinic.