Multicomponent variance estimation for binary traits in family‐based studies

Abstract

In biometrical genetic analyses of binary traits, the use of family data overcomes some limitations of twin studies, particularly in terms of sample size and types of genetic or environmental factors that can be estimated. However, because of computational problems, recent methods in the application of generalized linear mixed models for family data structure have limited the ability to handle large data sets with general covariates. In this paper, we investigate the use of the hierarchical likelihood approach to the analysis of binary traits from family data. In a simulation study, the method is shown to be highly accurate for the estimation of both the variance components and fixed regression parameters, even for small family sizes. For illustration, we analyze a real data set of familial aggregation of preeclampsia, a pregnancy-induced hypertension. When possible, the analysis is compared with the exact maximum likelihood approach. Genet. Epidemiol. 30: 37-47, 2006. r 2005 Wiley-Liss, Inc.

Score Test for Linkage in Generalized Linear Models

“…We refer the reader to www.mlwin.com for a list and review of possible softwares. Practical solutions appear to be problemspecific and a few authors have dealt with this problem in the case of family data [3][4][5] . Besides, in some instances (e.g., when sib-pair data only are available), the GLMM may lack identifiability.…”

“…Score Test for Linkage in GLM Hum Hered 2007;64: [5][6][7][8][9][10][11][12][13][14][15] where a 2 denotes the total additive genetic variance and c 2 the common-environment variance on the underlying random effect scale.…”

“…This unobserved latent process is linked to the actual trait values via a traditional Generalized Linear Model (see Section 2). In fact, this type of models has already been used for estimation of the heritability of binary traits [3][4][5] as well as for linkage of longitudinal continuous [6] data and survival data [7] , although appealing GLMMs are in general difficult to fit with family data. Besides, we favor simple mathematically tractable expressions for a test, this is to reduce computational burden, but even more importantly, because we would like to get insight into the properties of this model when used in linkage studies.…”

Extended Bayesian Model Averaging in Generalized Linear Mixed Models Applied to Schizophrenia Family Data

“…We assume that these two parental effects are independent, f 2 and R matrices represent the variance and correlation matrix, respectively. To define the correlation matrix of parental effects, the genetic correlation between full siblings is assumed to be 0.5, that between parents is zero, and that between parents and children is the same as that between full siblings (Pawitan et al, 2004;Noh et al, 2006). For instance, if the size of a family is 4 (say, both parents and two children), then the correlation matrix for the paternal and maternal effects is This simple model follows a strong assumption that only genetic effects exist in the familial clustering of schizophrenia.…”

“…Many current approaches therefore rely on asymptotic and/or Markov chain Monte Carlo (MCMC) methods for statistical inference. For instance, the penalised quasi-likelihood (PQL) and a bias correction for PQL have been proposed; although both were not completely bias free (Noh et al, 2006).…”