Finding Associations among Histone Modifications Using Sparse Partial Correlation Networks

Julia Lasserre, Ho-Ryun Chung, Martin Vingron

Abstract: Author Summary Nucleosomes are protein complexes around which the DNA is wrapped for compactness. They are made of histone proteins that can be post-translationally modified and these histone modifications can affect the expression of surrounding genes. In the past decade, scientists have developed a strong interest in the part of gene regulation provided by epigenetics, i.e. those heritable characteristics that are not based on the DNA sequence and that can therefore be cell-type-specific, such as histone modifications. Striking patterns about the co-occurrence of modifications have been discovered, leading to the hypothesis that different combinations of modifications lead to different effects. Different histone modifications could act jointly to recruit certain proteins, or be required sequentially, which is reflected in statistical dependencies in measured data. The focus of this article is on building a network that represents the global dependencies by extracting direct associations of histone modifications. We find that, although histone modifications patterns are cell-type specific (modifications may not necessarily appear at the same loci), the dependencies are to a large degree cell-type independent, which is supported by a large overlap of the inferred associations in the networks built for different cell types. We are able to find meaningful associations, both known and novel.