Negative control variables for testing conditional genetic associations
As part of the DEMS Statistics Webinar series Professor Matteo Sesia (University of Southern California, Marshall School of Business) will present the paper:
"Negative control variables for testing conditional genetic associations"
Wednesday, December 2, 18:00
Abstract:
Genome-wide association studies measure hundreds of thousands of genetic variants in large cohorts, aiming to discover biologically meaningful relationships with a phenotype of interest. The statistical analysis of these data is challenging because the problem is high-dimensional, the explanatory variables are not independent of each other, the unknown model linking genotypes with phenotypes may be complex, and different sources of confounding complicate the interpretation of any findings.
This talk will discuss recent methods that address these difficulties by testing the conditional independence of a phenotype with a variant, or a subset thereof. The main idea is to approximate the distribution of genotypes in a population with hidden Markov models, and then simulate synthetic variables to be used as negative controls.
These solutions control the false discovery rate (FDR) and produce results that are directly interpretable, sometimes even in a formal causal sense.