Mediation-Based Integrative Genomic Analysis

• Activity Number: 309
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Sheila Gaynor* and Xihong Lin
• Companies: Harvard University and Harvard School of Public Health
• Keywords: Genomics ; Statistical Genetics ; Integrative analysis ; Mediation

Abstract

Genetic association methods have traditionally been used to analyze the relationship between SNP or sequencing data and disease outcomes. This standard approach often fails to explain a significant proportion of disease and elucidate the complete relationship between SNPs and complex diseases. It has thus been suggested that studies may be improved by jointly analyzing SNP and gene expression data to analyze phenotypes of complex diseases. Huang, VanderWeele and Lin (2014) proposed that this can be approached using a mediation model, where the association between SNP sets or whole genome sequences and a disease outcome is mediated by gene expression or epigenetic data. We explore this framework for real data sets. We explore utilizing the whole genome SNP data, gene expression and methylation data, and clinical data on disease phenotypes such as histology. We suggest that the application of such a mediation framework can quantitatively and qualitatively supplement a traditional genetic association study by providing explanation of the mechanisms leading to disease phenotypes. Further, we provide empirical evidence suggesting for which genomic settings this framework may be useful.