Abstract:
Population scale datasets like the UK Biobank provide a rich data source for both the identification of targets and the prediction of on-target adverse effects (target liability). Such assessment of target liability rests on the idea that phenotypes arising from natural variation affecting protein coding genes will predict adverse effects as well as the primary action from modulating that protein. Here we evaluate the use of an integrated toolset intended for phenotype mining, sample and variant QC, genome wide association, LD clumping and annotation with curated and harmonized reference datasets. We propose a workflow to identify genetic evidence to validate an existing candidate drug target and predict on-target adverse effects. To illustrate the process, we use AChE inhibitors (used for the treatment of Alzheimer’s disease {AD}), which unfortunately have been associated with heart rhythm dysregulation. We selected case and control groups for AD and arrhythmia, each with 150k+ participants, and ran GWAS for the entire imputed dataset (97 million variants) in a matter of hours. Leveraging Ensembl VEP and GTEX eQTL and sQTL annotations integrated within the platform, we identified two variants significantly associated with AD that were ACHE eQTLs or sQTLs within a range of tissues, including brain frontal cortex and heart atrial appendage. Pre-populated yet modifiable queries allowed seamless cross referencing with the GTEX data set. Although these variants were not found to associate with arrhythmia as defined in our preliminary tests, this workflow allowed us to reach this conclusion in a matter of days. Future work will involve testing association with narrower heart rhythm dysregulation phenotypes and testing for association within specific subpopulations
