Nonparametric Estimation for Genetic Mixture Models with Random Effects

• Activity Number: 406
• Type: Topic Contributed
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Tanya Garcia*+ and Yuanjia Wang and Yanyuan Ma
• Companies: Texas A&M University and Columbia University and Texas A&M University
• Keywords: Huntington's Disease ; Censored data ; Mixture Model ; Complete statistic ; Random Effects

Abstract

We study genetic mixture models for Huntington's disease (HD) data from the Cooperative Huntington's Observational Research Trial (COHORT). Earlier analyses showed that early HD onset is linked to long CAG repeats in the HD gene. To flexibly model the relationship between CAG repeat length and HD onset, Ma and Wang (2013) developed a partially linear varying-coefficient model under the logit link function. Their model makes no parametric assumptions about the relationship between CAG repeat length and HD onset; and accommodates right-censoring of HD onset and that the participants' genotype statuses are unobserved. Though successful, their model ignores the dependency in the COHORT data as it is collected via family units. To remedy this, we incorporate random effects into the Ma and Wang (2013) model without imposing distributional assumptions on the random effects. We show that our new model has properties similar to those of a sufficient complete statistic. This yields semiparametric estimators that are evaluated using closed-form expressions and a local kernel and back-fitting procedure. Simulation results and application to the COHORT data demonstrate our method's usefulness.