Identifying chromatin interactions at high spatial resolution

Abstract

This thesis presents two computational approaches for identifying chromatin interactions at high spatial resolution from ChIA-PET data. We introduce SPROUT which is a hierarchical probabilistic model that discovers high confidence interactions between binding events that it accurately locates. We apply SPROUT to CTCF ChIA-PET data from mouse embryonic stem cells and demonstrate that SPROUT discovers interactions that are more consistently supported by biological replicates than an alternative method called The ChIA-PET Tool. We also introduce GERM which models genome-wide distributions of protein occupancy without assuming that proteins can be accurately modeled as binding to point locations. We demonstrate that the locations that GERM identifies as interacting with transcription start sites of genes accurately align with ChIP-Seq data that are associated with active enhancers. Finally, we apply GERM to RNA Polymerase II ChIA-PET data from embryonic stem cells and motor neuron progenitors and make several observations about the usage of enhancers during motor neuron development.