B1717
Title: A hierarchical hidden Markov random field model for peak calling across multiple Hi-C datasets
Authors: Yun Li - University of North Carolina, Chapel Hill (United States) [presenting]
Abstract: The constantly accumulating Hi-C data provide rich information for calling peaks across multiple tissue/cell types, experimental conditions, and/or cell differentiation stages. However, statistical models and computational tools are still in their infancy. Multiple factors, including sequencing depth and heterogeneity across Hi-C experiments, pose great challenges for the development of proper and efficient methods. We propose a peak caller based on a hierarchical hidden Markov random field (HHMRF) model to detect long range chromatin interactions from multiple Hi-C datasets. In addition to model the spatial dependency in the local neighborhood, HHMRF is able to model dependency across multiple Hi-C datasets, leading to further improved statistical power. We conducted comprehensive simulation studies, and showed that HHMRF model outperforms competing methods that ignore the dependency structure and call peaks separately in each individual Hi-C dataset. Next, we analyzed a real Hi-C dataset on human H1 embryonic stem cells and four H1 derived cells, and found that the cell-type-specific peaks identified by HHMRF show higher overlap with cell-type-specific epigenetic features and cell-type-specific gene expression, compared to those identified by competing methods. HHMRF model has the potential to unveil the structural basis of cell-type-specific transcription regulation mechanism.
Journal