Multivariate Mixture Distribution Model to Analyse Twin Data of Unknown Zygosity
From large cohort samples in human populations, twin pairs can usually be identified through date-of-birth or other identifiers, so that a genetic analysis can be performed. However, with this kind of ascertainment the zygosity for same-sex pairs is not known. The mixture distribution model has been shown to be reliable in partitioning the phenotypic variance of twins of unknown zygosity into genetic and environmental components [Neale (2003) Twin Res. 6: 235-239; Benyamin, et al. (2005) Behav Genet 35: 525-534]. Those studies were concerned only with a single trait analysis, where the standard errors of the estimates from the mixture distribution are larger than those of the conventional (known zygosity) model, especially for small to moderate heritability. Additional phenotypes and multivariate analysis have been suggested to provide additional zygosity classification to lower the standard errors of the estimates. Using computer simulations, this study quantifies the precision of the multivariate mixture distribution model compared with that of the univariate mixture model in analysing data from twins of unknown zygosity. It is shown that multivariate analysis decreased the variability (hence, the standard error) of the heritability estimates from the mixture distribution model. The more traits analysed simultaneously, the smaller the variability of the heritability estimates from the mixture distribution model. It was predicted that if more than 10 traits were analysed simultaneously, then the mixture distribution is as good as the known zygosity model with similar sample size. In conclusion, multiple correlated trait complexes can provide accurate estimates of genetic parameters from twin data of unknown zygosity.