A statistical model of ChIA-PET data for accurate detection of chromatin 3D interactions

Abstract

Identification of three-dimensional (3D) interactions between regulatory elements across the genome is crucial to unravel the complex regulatory machinery that orchestrates proliferation and differentiation of cells. ChIA-PET is a novel method to identify such interactions, where physical contacts between regions bound by a specific protein are quantified using next-generation sequencing. However, determining the significance of the observed interaction frequencies in such datasets is challenging, and few methods have been proposed. Despite the fact that regions that are close in linear genomic distance have a much higher tendency to interact by chance, no methods to date are capable of taking such dependency into account. Here, we propose a statistical model taking into account the genomic distance relationship, as well as the general propensity of anchors to be involved in contacts overall. Using both real and simulated data, we show that the previously proposed statistical test, based on Fisher's exact test, leads to invalid results when data are dependent on genomic distance. We also evaluate our method on previously validated cell-line specific and constitutive 3D interactions, and show that relevant interactions are significant, while avoiding over-estimating the significance of short nearby interactions.

Chromatin interaction analysis with updated ChIA-PET Tool (V3)

“…Interacting pairs 1,762,737 NA NA (10) Note: NA -Not applicable With the developing of ChIA-PET, there are more and more published tools such as, Mango [13], ChIA-PET2 [14] and ChiaSig [15] and so on, to process and analyze ChIA-PET data in recent years. In order to evaluate ChIA-PET Tool V3, we use ChIA-PET data associated with RNAP II from human K562 cells as input data and compare the results of ChIA-PET Tool V3 with other tools (Table 8).…”

An integrated 3-Dimensional Genome Modeling Engine for data-driven simulation of spatial genome organization

“…PET interactions were identified and classified following the approach in Li et al (2010), with modifications described by Tang et al (2015), using a PET cutoff of 4-PET clusters with counts <4 are considered not significant, and these weak interactions are grouped with PET singletons. Although we prefer this analysis method, several methods are available for calling PET interactions (Paulsen et al 2014;He et al 2015;Phanstiel et al 2015), and the output of any of these methods is suitable as input to 3D-GNOME.…”

Statistical Challenges in Analyzing Methylation and Long-Range Chromosomal Interaction Data

“…Recognizing this problem, Fullwood et al treated pairs that are connected only once as false pairs [112]. Simple tests such as hypergeometric (HG) and weighted (generalized) HG [139] are used to further filter out false loops. In the weighted HG test, data are “normalized” in the sense that pairs in close proximity in the 1D (linear) genome are treated as more likely to have random collisions.…”