|
Abstract:
|
The use of genome-wide testing thresholds in association scans is known to inflate estimates of genetic risk for significant SNPs (the winner's curse" or "significance bias"). We have recently reported an approximate conditional likelihood approach to correct for this bias using the estimated risk effect and its standard error. A similar problem arises when risk estimation is performed for secondary effects, such as secondary phenotypes or gene-environment interactions, when the secondary analysis is restricted to SNPs that are significant for the primary phenotype. For case-control samples, the proper analysis of secondary phenotypes may be nontrivial, and provides additional complexity to the estimation problem. We describe an extension of our conditional likelihood approach to the multivariate setting where multiple effect coefficients are simultaneously estimated.
|
