Efficient SNP-Based Heritability Estimation Using Gaussian Predictive Process Modeling in Large-Scale Cohort Studies

• Activity Number: 43 - Statistical Genetics II – New Models for Complex Study Designs
• Type: Contributed
• Date/Time: Monday, August 3, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): Souvik Seal* and Abhi Datta and Saonli Basu
• Companies: University of Minnesota and Johns Hopkins University and Division of Biostatistics, University of Minnesota
• Keywords: Heritability; Predictive Process; Genetic Relationship; UK Biobank; Dimension reduction; Fast optimization

Abstract

Genome-wide association studies have reported many SNPs associated with complex traits, but in aggregate they explain very little of the given trait’s heritability as estimated from twin and family studies. Recently classical methods for estimating heritability based on known pedigree relationships have been significantly augmented by using high-dimensional Genetic Relationship Matrix (GRM) constructed using genome-wide SNP data on nominally unrelated or distantly related individuals. Such likelihood-based methods usually involve huge matrix operations and thus, face significant instability and computational deadlock in large scale cohort studies (UK Biobank, Precision Medicine cohort, Millions Veterans Program are few such examples). In such a context, we propose a heritability estimation approach primarily based on Gaussian Predictive Process model. It involves matrix operations of much lower dimensions, thereby massively easing the computational burden. We estimate the heritability of several quantitative traits from the UK Biobank data with five hundred thousand individuals.