| Summary: | Though widely believed to exist, few convincing examples of genetic interactions have been detected through statistical approaches in genome-wide association studies. The first piece of work in this thesis attempts to determine if there is evidence for the existence of such interactions within genes identified through protein-protein interactions. A software package is developed and applied to data from a recent publically available genetic study. No evidence was found for an enrichment of such interactions in the available data. The second study applies three current methods for interaction detection to a real data set with compelling evidence of an interaction. Sparse Partitioning, SNPHarvester and Random Jungle were selected, with the later two being followed by the HyperLasso as a post-processing step. Only one method, SNPHarvester, was able to detect the interaction. The third study outlines a local pilot project in renal transplant dysfunction. Genetic variants from donors and recipients are examined using survival analysis. Interactions between the two genomes are tested for an effect on the survival time of the kidney. Secondary renal phenotypes of acute rejection and progression to end-stage renal failure are also considered. There were no strongly significant associations discovered in this data. The final study is a multi-centre renal transplant study analysing over 2000 donor recipient pairs throughout the UK and Ireland. Although much larger than the pilot, this study also failed to detect any associations with graft survival time or the secondary phenotypes. SNPHarvester was applied to the data and there are some indications of potential interactions, but replication is essential before the results can be trusted. An outline of an extension to SNPHarvester to better handle survival data is presented. Results from all of these studies were largely negative. Detecting interactions in genome-wide data remains a difficult task. Narrowing the search space by filtering may be a better approach than attempting a genome-wide search, though SNPHarvester has proven to be useful and is a good choice if a true genome-wide search is required.
|
|---|
