Developing and evaluating tools to improve the quality of DNA methylation association studies

Abstract

There is increasing interest in studying DNA methylation in the context of health and disease. A number of technical and analytical considerations are important to take into account when designing and interpreting DNA methylation studies, such as the experimental parameters used when quantifying DNA methylation differences between individuals and how best to account for study confounders, such as cellular composition. This thesis aims to address these issues by first developing a method to assess study power in bisulfite sequencing (BS) studies, second establishing a method for the estimation of error across reference based cellular deconvolution models, and third generating a novel reference based DNA methylation deconvolution model for the brain incorporating data for three neural cell types. In Chapter 2 the impact of bisulfite sequencing depth and sample size on power is investigated. It is shown that study power is not dependent on one specific parameter, but reflects the combination of multiple study-specific variables. Data simulation is utilised to generate an interactive tool for use by the wider research community that can be used to estimate the power of BS studies based on user-defined input variables including sample size and read depth filtering. In Chapter 3 an error metric is established for reference based cellular deconvolution approaches using DNA methylation data, which is validated using datasets derived from both blood and brain tissue. In Chapter 4 the reference based deconvolution model utilised for the deconvolution of brain tissue is refined to include an additional cell type, resulting in a three cell type model. The model was applied to bulk brain DNA methylation samples, showing that the addition of a third cell type improved insight gained from data generated on bulk brain tissue. Overall, this thesis aims to generate tools which can be utilised to better design and interpret DNA methylation studies, all of which have been made publicly available. This thesis also encourages researchers to clearly communicate any DNA methylation quality control decisions made and examine their methodologies to improve the transparency and reproducibility of their findings.