|
Abstract:
|
The Collaborative Cross (CC) is an exciting resource for geneticists as it mimics the genetic variability of human populations. Studies involving recombinant inbred crosses (RIX) of CC lines provide an interesting application for mixed models, by modeling parental strain effects as random. The complicated correlation structure in the data poses challenges for categorical phenotypes. Inference on variance components (itself a "nonstandard" problem as parameters of interest fall on the boundary under the null) is not straightforward, as the response vector cannot be divided into independent units. We show that a score statistic developed by Lin (1997), which is applicable when fitting by pseudolikelihood, is valid in our more general shared random effect setting. This statistic is particularly useful in trait mapping, as the model need only be fit under the null. We apply the statistic to data collected on RIX crosses of 63 CC lines that investigated the effect of haloperidol on phenotypes measured in an open field experiment, and on adverse events. For all outcomes, we identify strong strain effects, and many strain effects that interact with treatment.
|
