BIRS Workshop Lecture Videos
Heteroskedastic linear models for functional genomics Engelhardt, Barbara
Heteroskedasticity in linear models refers to the situation where a predictor -- here, a single nucleotide polymorphism (SNP) -- is correlated with the residual error of a linear model fitted to that SNP and a quantitative trait response. Previous work in functional genomics has identified examples of a genotype affecting not (only) the mean of the quantitative trait, but (also) its variance, implying a heteroskedastic association. Methods to identify these variance QTLs (vQTLs) are based on ANOVA tests or two-stage linear models. Here, we develop a test for heteroskedasticity based on a Bayesian heteroskedastic linear model. We show the power of this test for identifying vQTLs for various cellular traits. We extend the model to pairs of quantitative traits to address questions about causal relationships among cellular phenotypes.
Item Citations and Data
Attribution-NonCommercial-NoDerivatives 4.0 International