Alterations in epigenetic control have been linked to numerous age-related human pathologic conditions including cancers and cardiovascular disease, however the mechanisms by which these alterations occur are not well understood. Numerous studies have reported deceasing methylation throughout the aging process. Furthermore, recent work has demonstrated an association between tumor progression and alterations in methylation profiles which parallel the changes seen with age. Although such alterations vary considerably between cancer types, it is thought that abnormal methylation contributes to the progressive inactivation of tumor suppressor genes during aging, which subsequently results in growth advantage toward cancer development. Hence, the aging process and differences in environment have been hypothesized to contribute to clinically significant alterations in epigenetic profiles as individuals accrue varying exposures with age. The proposed project seeks to understand the relationship between aging and DNA methylation and its role in the development and progression of age-defining illnesses. To do so, we formulate computationally efficient model-based clustering algorithms which incorporate complex covariance structures. Studies examining methylation patterns throughout the aging process are often longitudinal in nature. Thus, to fully characterize the impact of aging on methylation, methods that handle repeated measures data are necessary. To this end, the first aim of the proposed project involves the construction of a framework for a model-based clustering algorithm called Recursively Partitioned Mixture Models (RPMM) that incorporates covariance structures which characterize the longitudinal features of these data. Furthermore, it has been well documented that genes do not function independently of one another, but rather in complex biological networks. Moreover, CpG loci residing within the same gene tend to exhibit much stronger correlation patterns compared to CpG loci across genes. Consequently, our second and third aim seek to provide a framework for RPMM that integrates covariance structures which make use of these features. Doing so more accurately reflects the biology involved, which will lead to an enhanced understanding of DNA methylation and its role in the development and progression of age-defining illnesses.
The novel contribution of this project is the development of model-based learning algorithms that incorporate complex covariance structures. The broader impact of this project include: (1) an enhanced understanding of the relationship between epigenetics and aging and (2) the role of epigenetics in the development and progression of age-defining illnesses. The proposed project is significant because the covariance structures we propose reflect the true nature of methylation data, which will lead to an enhanced understanding of the epigenetics of aging as well as the role of epigenetics in the development of age-defining illnesses.
|Koestler, Devin C; Marsit, Carmen J; Christensen, Brock C et al. (2014) A recursively partitioned mixture model for clustering time-course gene expression data. Transl Cancer Res 3:217-232|
|Koestler, Devin C; Christensen, Brock C; Marsit, Carmen J et al. (2013) Recursively partitioned mixture model clustering of DNA methylation data using biologically informed correlation structures. Stat Appl Genet Mol Biol 12:225-40|