RATTACA: Genetic predictions in Heterogeneous Stock rats offer a new tool for genetic correlation and experimental design

Abstract

Genetic correlations between traits are a common first step in studies identifying causal genetic pathways and mechanisms. Using this framework with inbred or selected lines, however, requires intensive labor investment through breeding and phenotyping, and is prone to confounding, as observed trait correlations do not necessarily reflect a causative genetic architecture shared between distinct populations. When drawn from a single outbred population, genetic trait predictions offer a viable alternative to experimental phenotyping and can be used to identify putative genetic correlations when samples with divergent trait predictions also diverge in a second measured trait. Here, we present a novel research paradigm and service called RATTACA, in which genotypes from Heterogenous Stock (HS) rats are used to predict trait values using linear mixed models. These predictions are used to select samples of individuals with high and low extreme trait values, facilitating (1) a priori sampling of desired trait values without oversampling across phenotypic space and (2) easy identification of putative genetic correlations between predicted and newly measured traits. We validated prediction models using four example phenotypes with measured trait values and found sufficient accuracy to distinguish extreme trait samples, even when using a small number of genome-wide variants (n = 50,000) for traits with modest heritability (h2 = 0.13). Given genotypes and trait measurements available through previous research in HS rats, we propose RATTACA as a service to reliably predict more than 80 behavioral and physiological traits.