A power-based sliding window approach to evaluate the clinical impact of rare genetic variants

Abstract

Systematic determination of rare and novel variant pathogenicity remains a major challenge, even when there is an established association between a gene and phenotype. Here we present Power Window (PW), a novel sliding window technique that identifies the clinically impactful regions of a gene using population-scale clinico-genomic datasets. By sizing windows based on the number of variant carriers, rather than the number of variants or nucleotides, statistical power is held constant during analysis, enabling the localization of clinical impact as well as the removal of unassociated gene regions. This method can be used to focus on: specific variant types such as loss of function (LoF) or other coding; parts of a gene, such as those expressed in different tissues; or isolating gene regions with opposite directions of effect. Using a training set of 300K exomes from the UKBiobank (UKB), we developed PW-based LoF and coding models for well-established gene-disease associations and tested their accuracy in two additional cohorts (128k exomes from the UKB and 30k exomes from the Healthy Nevada Project (HNP)). The significant PW models retained a mean of 64% of the rare variant carriers in each gene (range 16-98%), with quantitative traits showing a mean effect size improvement of 48% compared to aggregating rare variants across the entire gene, and the odds ratios for binary traits improving by a mean of 2.4-fold. PW showcases that EHR-based statistical analyses can accurately distinguish between novel coding variants that will have high phenotypic penetrance in a population and those that will not, unlocking new potential for population genetic screening.

Withdrawn: Hypertension increases PPV for polycystic kidney disease inPKD1andPKD2variant carriers

“…Variants were classified as LoF if the consequence was stop_lost, start_lost, splice_donor_variant, frameshift_variant, splice_acceptor_variant, or stop_gained. Additionally, LoF variants were filtered to MAF < 0.1% in each gnomAD and local UKB or Helix population and, based on our prior study of these genes and phenotype with the Power Window method (21), fewer than 20 carriers in UKB. Any individual with a variant meeting this criteria was determined to be a carrier of either a PKD1 or PKD2 LoF variant, and no individuals in our study had a qualifying variant in both genes.…”

Patients with Type 2 Diabetes and aGCKvariant are still at risk for T2D-related secondary complications

“…We found that coding variants that increase glucose levels above the median values of those without GCK variants (89 mg/dL) could be reliably identified if: (i) the statistical evidence map predicted the variant to increase glucose levels and (ii) the functional evidence map did not predict the variant to decrease glucose levels. Those with GCK variants that did not meet this criteria, on average, displayed glucose levels in line with those without GCK variants ( Figure 1B ) 26,27 .…”

“…In this assay, resulting GCK enzymatic activity levels can be quantified by the growth rate of yeast cells carrying each variant on a glucose-based medium 25 . The second dataset is a statistical evidence map, which evaluates association patterns with glucose levels across the gene using clinico-genomic data, processed by creating statistical power-based windows of rare variants of relevant molecular consequences across a population 26 . We built the GCK statistical evidence map with genetic data from 333,190 randomly selected UKB participants, using random blood glucose (i.e., not necessarily measured after fasting) collected as part of their baseline assessment visit as our phenotype ( Figure 1A ).…”

“…For coding variants classified as missense_variant, inframe_deletion, or inframe_insertion, non-benign (by Polyphen or SIFT), and MAF<0.1%, we also used a combination of a functional evidence map (urn:mavedb:00000096-a), with scores for each variant assigned using a yeast complementation assay, and a statistical evidence map created using the Power Window technique to predict variants that likely increased glucose levels 25,26,39,40 . Briefly, Power Window is a sliding window analysis that groups variants located near each other into one unit and analyzes them together to improve power, much like a gene-based collapsing analysis but at a smaller scale, as previously described 26 . We trained the Power Window model for GCK on ~330k individuals (n= 333,190) from the UKB, used a beta cutoff of 0.5 to build the model, and tested on ~112k (n=112,015) unrelated UKB individuals ( Figure 1A,B ).…”

“…Variants were classified as LoF if the consequence was stop_lost, start_lost, splice_donor_variant, frameshift_variant, splice_acceptor_variant, or stop_gained and MAF<0.1% in all ancestry populations in gnomAD and the analyzed cohorts. For coding variants classified as missense_variant, inframe_deletion, or inframe_insertion, non-benign (by Polyphen or SIFT), and MAF<0.1%, we also used a combination of a functional evidence map (urn:mavedb:00000096-a), with scores for each variant assigned using a yeast complementation assay, and a statistical evidence map created using the Power Window technique to predict variants that likely increased glucose levels 25,26,39,40 . Briefly, Power Window is a sliding window analysis that groups variants located near each other into one unit and analyzes them together to improve power, much like a gene-based collapsing analysis but at a smaller scale, as previously described 26 .…”