Discover The Semantic Structure Of Human Reference Epigenome By Differential Latent Dirichlet Allocation

Understanding epigenetic changes across various conditions is a fundamental problem to epigenome annotation. With more high-throughput epigenomic data available, computational methods have been developed to quantify various types of epigenetic modification signals, to compare epigenetic marks between different conditions and to understand the functional consequences of epigenetic changes. However, currently few studies on epigenomes aim to provide a global view of epigenetic changes through large-scale high-throughput data integration. We apply a probabilistic graphical model called differential latent Dirichlet allocation (DLDA) to discover latent epigenetic modification modules (EMMs) from 56 reference epigenomes. We demonstrate the identified EMMs can characterize the global semantic structure of the reference epigenomes. The resulted EMMs show their condition-relevance to the corresponding reference epigenomes. The genes involved in these EMMs show epigenome-relevant functionality. Study of the involved epigenetic modification marks involved in these EMMs reveals the relative activity levels of epigenetic marks in different epigenomes. Clustering gene-epigenetic modification pairs leads to the discovery of more functional epigenetic modification groups