P2-9 GWAS Meta-Analysis for Pathways and Set Evaluations
PhD student: Stefanie Friedrichs
Thesis Committee: Heike Bickeböller, Tim Beißbarth, Thomas Kneib
Genome-wide association studies (GWAS) examine the associations between genetic variants and traits of interest across the genome. Studies of this type have already identified thousands of single nucleotide polymorphisms (SNPs) for complex diseases like cancer.
Despite these findings and ongoing improvements in the methods, a large part of the heritability still cannot be explained. This is partly due to the fact, that a person's susceptibility to complex diseases is influenced by the combined effects of different genes, environmental factors and their complex interactions in pathways. Pathways are networks, consisting of interacting genes, which represent key biological functions of the human organism. Therefore pathway-based analyses try to find associations not only of a single marker but for a combination of multiple genetic markers belonging to a known network.
The logistic kernel machine test is a powerful tool for pathway analysis. The size-invariant kernel can account for different numbers of SNPs and genes within a pathway and the network kernel even enables to incorporate prior knowledge of the network structure connecting a pathways genes.
However, both new kernels still only give a test result for the whole pathway and a way to detect causal genes within the identified pathway is still missing. I am interested in finding a way to yield significances and effect sizes on a gene-level as well.
Combining different GWAS studies in meta-analysis is essential to assess the overall evidence given by the participating studies. Joint analysis of results from independent studies investigating the same disease can increase the power to detect associations.
Normally in a meta-analysis we would consider the estimated effect size with its confidence interval, the p-value, sample size and the direction of the effect. Testing pathways, we can only obtain a p-value which considerably limits the applicable methods.
Not knowing the direction of a detected association can lead to a problem when combining p-values, as effects in opposite directions will be regarded as findings confirming each other although they are contradictory. This might lead to an increase of false positives. The situation gets particularly challenging in the context of pathways where multiple SNPs with effects in varying directions are considered and additionally those directions are unknown. I would like to find an appropriate method to perform a meta-analysis on pathway-based results.
back to overview of all projects