|
Abstract:
|
Case-parent triad studies sample affected offspring, manifesting a disease or phenotype of interest, as well as their parents. If parents of the offspring are alive and consent to genotyping, the use of nuclear families can be a powerful way to measure association. For some triads, not all genotypes are available for each family member, usually the paternal DNA, at the time of the study. The EM algorithm is widely used to impute missing genotypes. Some of these algorithms under-estimate the variance of the regression coefficients, especially when the minor allele frequency is less than 0.10 or when the degree of missingness is large (> 20%). I propose a new method by solving a Zero-Inflated Poisson (ZIP) likelihood directly. This likelihood simultaneously estimates allele frequencies, regression coefficients and imputes missing data to reduce bias and type I error, and maintain power.
|
