Function-on-Scalar Regression for Next-Generation Sequencing Studies

• Activity Number: 248
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Olga Vsevolozhskaya* and Dmitri Zaykin and Qing Lu
• Companies: and National Institute of Environmental Health Sciences and Michigan State University
• Keywords: rare variants ; common variants ; association test ; qualitative trait ; quantitative train ; functional analysis

Abstract

To improve the power of detecting an association, it is often desirable to link genotypes of individuals with several related phenotypes. For genetic association studies with multiple phenotypes, a new strategy is proposed based on function-on-scalar regression models. Envision plotting genotype values of genetic variants on the y-axis versus the variants' positions on the x-axes and 'connecting the dots'. Conceptualized in this way, the stream of genotype values can be visualized as continuous curves smoothly varying over genetic loci and treated as functional responses in a regression model. Multiple phenotypes can be added to the right-hand side of the equation as scalar covariates, bearing continuous effects smoothly changing over variants' positions. Whereas genetic information is typically added as covariates into the model, treating genotype values as functional responses provides some advantages: it simplifies the choice of a link function and alleviates the burden of multiple testing. By exploiting a connection between penalized splines and BLUPs, our proposed model can be fit using standard linear mixed model software, which facilitates practical implementation.