Abstract

Epigenetic modifications such as histone modifications and DNA methylation play a central role in the regulation of gene expression. Next generation sequencing technologies are now enabling genome-wide measurements of epigenetic marks, yet the data returned by such technologies is difficult to interpret. Here, we start from the observation that such measurements often return broad, spatially correlated patterns of modification which are highly reproducible between replicate experiments. We then use machine learning methodologies to exploit such spatial correlations to define stronger prediction methods. In particular, I will illustrate novel statistical hypothesis testing methodologies for ChIP-Seq (MMDiff) and BS-Seq (M3D) data, which exploit spatial features to yield more powerful tests. I will also elaborate on what may be the mechanisms underpinning the presence of such spatial correlations, and illustrate how higher-order spatial features may be used to predict gene expression from DNA methylation alone.