Browsing by Author "Tuominen, Samuli"

Modern day technology and computational power have allowed a large scale investigation of the human epigenome. Out of the epigenetic modifications, DNA methylation is of particular interest, since it is relatively easy to measure and very common in the DNA. A methylation site is a region of the DNA sequence that shows variation in the DNA methylation between individuals. Epigenome-wide association studies (EWAS) examine the interaction between these methylation sites one at a time and a specific human trait or an enviromental exposure. EWAS studies are, however, limited by low statistical power and problems related to multiple testing. To counter these issues, polygenic methylation scores have been developed to aggregate information over many methylation sites. These scores have two main applications. First is to formulate new hypotheses to explain human trait variation. Second one is to indicate unobserved environmental factors in cohort based studies or to predict individual developmental or disorder related outcomes. At the beginning of this thesis there is an introduction to epigenetics, to EWAS and polygenic methylation scores and to their genetic counterparts, genome-wide association studies (GWAS) and polygenic risk scores (PRS). Much of the methodology relating to the methylation scores is borrowed from GWAS and PRS. Some statistical properties of the methylation scores are derived in this thesis with focus on how the statistical power of detecting true association between a phenotype and human DNA methylation depends on the make up of the methylation scores. The theoretical derivations are tested through simulations. This thesis also examines how methylation scores may be calculated in practice using cross-validation and correlation reduction procedure called clumping. The methodology is applied to a Finnish cohort from the prediction and prevention of preeclampsia and intrauterine growth restriction study (Predo). The comparison of theoretical and observed statistical power in the simulations show that the theoretical and observed power correspond well to each other. In the practical analyses conducted using the DNA methylation data set and phenotype data of the Predo cohort and a maternal body-mass index (BMI) EWAS data, a clear piece of evidence of association of maternal pre-pregnancy BMI and offspring DNA methylation is found. The results support the growing evidence for the applicability of methylation scores in indicating prenatal environmental factors from the DNA methylation of the offspring.