Investigating genetic architecture of extreme and unexpected phenotypes using phenotype prediction in the UK Biobank (360G-Wellcome-224890_Z_21_Z)

Genetics underpins variability in observed traits. While many people study the genetic effects driving phenotypic variation for the general population, I am interested in using data from the UK Biobank1 to study the genetic effects responsible for the rare individuals with outlier phenotype measurements. I will consider two types of phenotypic outliers: 1\. An individual whose measured trait is at an extreme end of the cohort-wide distribution. 2\. An individual whose measured trait differs significantly from the trait expected based on their genetic predisposition. To identify this type of outlier we start by predicting the expected phenotypes. We will use both existing methods for phenotype prediction using genetics or family history of disease, and develop a novel method trained on data from the ALSPAC dataset7. Using these definitions, we will determine outliers for all heritable and commonly measured phenotypes across ancestry groups. We will seek to discover associations between outlier status and genetic variation, outlier status for other phenotypes, sex, age, and ancestry. I expect this work to improve our understanding of the genetic basis for outliers, lead to an understanding of ancestry-dependent diagnosis patterns, explain why some existing phenotype prediction methods perform poorly, and inform future phenotype prediction methods.