Predicting Phenotypes of Arbitrary Related Individuals Using Penalized Maximum Likelihood Method

• Activity Number: 127
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Xuesong Li*+ and Lan Zhu
• Companies: Oklahoma State University and Oklahoma State University
• Keywords: phenotype ; prediction ; mixed model ; high dimension

Abstract

Predicting quantitative traits from genomic data provides considerable information in studies of human diseases and livestock breeding. However, phenotype prediction is usually confronted with genetic relatedness among individuals. It is even more challenging when there are quite a number of factors associated with traits under study (the problem of p > n). In this study, we propose to incorporate pedigree structure as a polygenic random effect into a mixed linear model to predict phenotypes. The beauty of our approach is that we can include all available markers in the prediction model, even in the situation of p > n. A penalized maximum likelihood method is derived to estimate parameters in the mixed model. The performance of predictions is evaluated through 95% coverage rate and mean squared prediction error. We also investigate the robustness of our proposed prediction model in some common scenarios. Specifically, we explore the effect of several main factors on the performance of our method. These factors include sample size, trait heritability, magnitude of additive effects, and complexity of pedigree structures.