Transcriptional decomposition reveals active chromatin architectures and cell specific regulatory interactions
Gene transcription is influenced by favourable chromosome positioning and chromatin architectures bringing regulatory elements in close proximity. However, it is unclear to what extent transcription is attributable to topological organisation and what fraction of expression is determined by gene-specific regulatory programs. Here, we develop a strategy to decompose expression data into two main compo- nents reflecting the positional relationship of neighbouring transcription units and effects independent from their positioning.
We demonstrate that the positionally dependent component is highly informative of topological domain activity and organisation, revealing boundaries and chromatin compartments. Furthermore, features derived from transcriptional components can accurately predict individual chromatin interactions and are, in particular, capable of inferring enhancer-promoter interactions. Strength of transcription and lower levels of positionally dependent transcriptional noise were strong predictors of interactions at all distances. On the other hand, enhancer strength as measured by detected eRNAs appeared important for long-range enhancer-promoter communications. Finally, we assess commonalities and differences in regulatory organisations across 76 human cell types. We observe extensive sharing of expression associated domain structures across cells but that enhancer-promoter interactions are highly cell-type specific. In all, we demonstrate a close relationship between transcription and topological chromatin architecture and provide an unprecedented resource for investigations of regulatory organisations across cell types.