Seminar Details

Seminar Details


Jun 25

1:00 pm

Models and Inference of Transmission of DNA Methylation Patterns in Mammalian Somatic Cells

Audrey Fu

Final Exam

University of Washington - Department of Statistics

DNA methylation is critical for normal development of humans and many other organisms. Aberrant methylation patterns can lead to developmental diseases and cancer. Here we study the transmission process of methylation patterns over somatic cell division in mammals, with the goal of understanding (1) to what extent methylation patterns are transmitted faithfully from one cell generation to the next, and (2) how processivity of methylation enzymes helps in shaping the methylation patterns along DNA sequence.

We develop statistical models and inference methods for double-stranded methylation data, which are pairs of binary strings. To tackle the first question, we estimate the rates of methylation events and assess the variability in the rates across sites, formulating the question as a latent variable problem and developing multi-site models to infer the latent strand type. These models, however, assume independence of events across sites. To study the spatial pattern and infer the level of processivity in enzymes, we model each type of methylation events as a hidden Markov chain. We further incorporate physical distances into the model and use sojourn times (or distances in our case) as a measure of the level of processivity. For both questions, our models can easily incorporate and estimate experimental error rates. Markov chain Monte Carlo techniques under a Bayesian framework are used for inference.

We analyze data collected at the promoter region of the {\it FMR1} locus on the hypermethylated X chromosome in females and compare our estimates to existing results.