Development of a statistical method for the identification of gene-environment interactions

Abstract

In order to understand common, complex disease it is necessary to consider not just genetic risks and environmental risks, but also the interplay between them. This thesis aims to develop methodology for the detection of gene-environment interactions specifically; both by looking at the strengths and weaknesses of traditional approaches and through the development and testing of a novel statistical method. Developments in genotyping technology enable researchers to collect large volumes of polymorphisms in human genes, yet very few statistical methods are able to handle the volume, variation and complexity of this data, especially in combination with environmental risk factors. Interactions between genes and the environment are often subject to the curse of dimensionality, with each new variable increasing the potential number of interactions exponentially, leading to low power and a high false positive rate. The Mixed Tree Method (MTM) exploits the differences between environmental and genetic variables, by selecting the most appropriate features from conventional methods (including recursive partitioning, random forests and logistic regression) and combining them with new comparison algorithms which rank the genetic variables by the likelihood that they interact with the environmental variable under study. Results show the MTM to be as effective as the most successful current method for identification of interactions, but maintaining a much lower false positive rate and computational burden. As the number of SNPs in the dataset increases, the success of MTM compared to other methods becomes greater while the comparator approaches exhibit computational problems and rapidly increasing processing times. The MTM is also applied to a colorectal cancer dataset to show its use in a practical setting. The results together suggest that MTM could be a useful strategy for identifying gene environment interactions in future studies into complex disease.