Paradigm of Genetic Risk Prediction: Variable Selection or Mixed Model

• Activity Number: 589
• Type: Topic Contributed
• Date/Time: Wednesday, August 7, 2013 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Peng Wei*+
• Companies: Univeristy of Texas School of Public Health
• Keywords: risk prediction ; genetics ; genomics ; GWAS ; variable selection ; mixed model

Abstract

Recent advances in high-throughput array and next-generation sequencing technologies, such as high-density SNP arrays, whole-exome and whole-genome sequencing, have enabled genome-wide investigation of complex human diseases. Besides discovering genes associated with disease risk, a substantial interest in genetics research is to build risk prediction models based on genetic and genomic data that can be used for predicting disease prognosis or to identify individuals at high risk of developing the disease. Most existing genetic risk prediction models only focus on a few genome-wide significant genes/SNPs, leading to moderate predictive performance. On the other hand, recently developed polygenic theory for complex disease inheritance suggests that there are hundreds or even thousands of disease-predisposing genetic loci but only with small to moderate effect size. Genetic risk prediction has thus been approached from two different perspectives: variable selection assuming sparse signals and mixed models assuming many weak signals. In this talk, I will compare these two paradigms and illustrate their connections using simulations and real GWAS datasets.