Imprinting and maternal effect detection using partial likelihood based on discordant sibpair data

Numerous statistical methods have been developed to explore genomic imprinting and maternal effects, which are causes of parent-of-origin patterns in complex human diseases and are confounded. However, most of these methods have limitations: they may model only one of the two confounded epigenetic effects; they may make strong, yet unrealistic assumptions about the population to avoid over-parameterization; or they are applicable only to study designs that require the recruitment of difficult-to-obtain control families. In this study, we develop a partial likelihood method for detecting imprinting and maternal effects for a discordant sibpair design (LIMEDSP ) utilizing all available sibship data without the need to recruit separate control families. By matching affected and unaffected probands and stratifying according to their familial genotypes, a partial likelihood component free of nuisance parameters can be extracted from the full likelihood. This alleviates the need to make assumptions about the population. Our theoretical analysis shows that the partial maximum likelihood estimators based on LIMEDSP are consistent and asymptotically normally distributed. Using a closed-form formula, we compare a study design with more independent families and a design with larger families by keeping the total number of individuals that need to be genotyped fixed. We also conduct a simulation study to demonstrate the robust property of LIMEDSP and show that it is a powerful approach that does not require recruiting control families. To illustrate its practical utility, LIMEDSP is applied to a clubfoot disease data set and to the data from the Framingham Heart Study.