Deep Learning-based Phenotype Imputation on Population-scale Biobank Data Increases Genetic Discoveries

Abstract

Biobanks that collect deep phenotypic and genomic data across large numbers of individuals have emerged as a key resource for human genetic research. However, phenotypes acquired as part of Biobanks are often missing across many individuals, limiting the utility of these datasets. The ability to accurately impute or "fill-in" missing phenotypes is critical to harness the power of population-scale Biobank datasets. We propose AutoComplete, a deep learning-based imputation method which can accurately impute missing phenotypes in population-scale Biobank datasets. When applied to collections of phenotypes measured across 300K individuals from the UK Biobank, AutoComplete improved imputation accuracy over existing methods (average improvement in r2 of 18% for all phenotypes and 42% for binary phenotypes). We explored the utility of phenotype imputation for improving the power of genome-wide association studies (GWAS) by applying our method to a group of five clinically relevant traits with an average missigness rate of 83% (67% to 94%) leading to an an increase in effective sample size of 2-fold on average (0.5 to 3.3-fold across the phenotypes). GWAS on the resulting imputed phenotypes led to an increase in the total number of loci significantly associated to the traits from four to 129. Our results demonstrate the utility of deep-learning based imputation to increase power for genetic discoveries in existing biobank data sets.

Phenotype integration improves power and preserves specificity in biobank-based genetic studies of MDD

“…Finally, we applied a new deep-learning imputation method, AutoComplete 28 , to the same phenotype matrix ( Methods ). AutoComplete improved estimated imputation accuracy for most phenotypes with >10% missingness (29/42), and increased average estimated R 2 by 2.9%.…”

Using GWAS summary data to impute traits for genotyped individuals

“… 34 , 35 However, since PRS-CS, as many other model-based PRS methods, assumes linear effects of some specific SNPs on a trait, its imputed trait values are based on the estimated linear effects of the selected SNPs (which may not be truly causal or associated ones) and accordingly are not suitable for subsequent association analyses, though they are useful for prediction. Relatedly, although other methods have been proposed to impute a few missing values of a focal trait using other traits, 37 , 38 , 39 they are not suitable for our purpose of large-scale trait imputation for downstream genetic association analysis because of the loss of specificity: by definition, any genetic variants associated with a trait used to impute the focal trait are expected to be associated with the imputed focal trait, even not truly associated with the (observed) focal trait.…”

Personalized Mood Prediction from Patterns of Behavior Collected with Smartphones

“…The STAND eligibility criteria and treatment protocol are described extensively elsewhere 29 . Briefly, participants are initially assessed using the Computerized Adaptive Testing Depression Inventory 31 (CAT-DI), an online adaptive tool that offers validated assessments of depression severity (measured on a 0-100 scale). After the initial assessment, participants are routed to appropriate treatment resources depending on depression severity: those with mild (35 ≤ CAT-DI < 65) to moderate (65 ≤ CAT-DI < 75) depression at baseline received online support with or without peer coaching 30 while those with severe depression (CAT-DI ≥ 75) received in-person care from a clinician (Materials and Methods).…”

“…In total, we obtained 1,325 features. Missing daily feature values (Sup Figure 4) were imputed using two different imputation methods, AutoComplete 31 and softImpute 32 (Materials and Methods), resulting in 29,254 days of logging events across all individuals.…”