Data-driven selection and parameter estimation for DNA methylation mathematical models.

Highlights • Aberrant DNA methylation patterns are a hallmark of diseases, such as cancer, Alzheimer's disease, and cardiovascular disease, and mathematical analysis of the mechanics of DNA methylation has the potential to give new insights into these mechanisms. • We develop a fast and robust methodology for selecting computational models for DNA methylation. • We demonstrate the effectiveness of the method on recovering parameter values associated with the processes which underpin DNA methylation. • Our work opens the possibility for further biological experimentation based on our finding that the parameters which govern DNA methylation are influenced by gene promoter methylation levels. • Our work showcases the utility of Bayesian analysis and its application to complex biological problems. Abstract Epigenetics is coming to the fore as a key process which underpins health. In particular emerging experimental evidence has associated alterations to DNA methylation status with healthspan and aging. Mammalian DNA methylation status is maintained by an intricate array of biochemical and molecular processes. It can be argued changes to these fundamental cellular processes ultimately drive the formation of aberrant DNA methylation patterns, which are a hallmark of diseases, such as cancer, Alzheimer's disease and cardiovascular disease. In recent years mathematical models have been used as effective tools to help advance our understanding of the dynamics which underpin DNA methylation. In this paper we present linear and nonlinear models which encapsulate the dynamics of the molecular mechanisms which define DNA methylation. Applying a recently developed Bayesian algorithm for parameter estimation and model selection, we are able to estimate distributions of parameters which include nominal parameter values. Using limited noisy observations, the method also identified which methylation model the observations originated from, signaling that our method has practical applications in identifying what models best match the biological data for DNA methylation. [ABSTRACT FROM AUTHOR]