Cell-type-specific variation in chromosome conformation may be related to biological function. However, the specific loci that drive this variation have not been identified. Metrics for comparing Hi-C experiments either calculate global measures of similarity between experiments or identify differential pairwise contact frequencies. However, there is currently no method for identifying locus-specific changes in localization. We developed a method, MultiMDS, to simultaneously infer and align 3D structures inferred from two Hi-C datasets. The output of the method is two aligned 3D structures, which are used to calculate locus-specific changes between the datasets. By applying MultiMDS to Hi-C data from GM12878 and K562, we quantified the degree to which cell-type-specific structural changes occurred along the A/B compartment axis, representing the nuclear interior/nuclear periphery axis. On average 42% of changes occurred along this axis (compared to 33% expected by chance), demonstrating the importance of lamina-associated repression in cell type identity. Focusing on changes within the A compartment, we found that loci that undergo intra-compartment changes were enriched for enhancers in one or both cell types. For loci with a cell-type-specific enhancer, we found that loci in the cell type that lacks the enhancer gain interactions with loci enriched for polycomb repression.

Topologically associating domains (TADs) are essential in constraining the activity of transcriptional regulatory elements. Previous studies have observed that changes in TADs structures are associated with altered transcriptional outcome, suggesting that architectural changes may play an important role in regulating gene expression. Identification of differential TADs structures across conditions will provide insights on condition-specific regulatory mechanisms, helping identify potential pharmacologic targets. However, so far little work has been done on detecting differential TADs structures.

Here we present a novel statistical method that can accurately and quickly uncover differential TADs structures from Hi-C data. Through a systematic evaluation, we show that our method outperforms the existing methods. To validate the identifications, we applied our method to a Hi-C dataset obtained from a knockout experiment that depleted a critical transcription regulator that co-localizes with CTCF, and identified the changes in TADs structures between the wild type and the knockout. Our results show that the identified differential TADs structures correspond well with the depleted sites, confirming the biological relevance of our identifications. We further used our methods to study the differentiation of TADs and its relationship with gene expression for two cell lines in the hemopoiesis lineage. It reveals interesting biological insights.

Deciphering the detailed mechanisms governing gene regulation often requires 3-D reconstruction of the chromatin fold. Chromosome conformation capture techniques such as 3C/Hi-C provide insight into the chromatin folding state but may only give averaged spatial information in the form of 2-D contact heat maps. Using these techniques alone, it is difficult to answer important questions such as: quantification of functional cellular sub-populations, identification of the cause-effect relationships among the spatially interacting genomic regions, and discovery of driver interactions governing chromosome folding. We present CHROMATIX (CHROMATin mIXture), a Bayesian method based on a physical model of chromatin folding which provides probabilistic answers to these questions. We demonstrate our method on an example TAD locus.

Phenotypic heritability of complex traits and diseases is seldom explained by individual genetic variants. Many methods have been developed to select a subset of variant loci, which are associated with or predictive of the phenotype. Selecting SNPs that are close on SNP-SNP networks have been proven successful in finding biologically interpretable and predictive SNPs. However, we argue that the connectedness constraint favors selecting redundant features that affect similar biological processes and therefore does not necessarily yield better predictive performance. In this paper, we propose a novel method called SPADIS that favors the selection of remotely located SNPs that are associated with the phenotype. This is achieved by maximizing a submodular set function with a greedy algorithm that ensures a constant factor (1 − 1/e) approximation. We compare SPADIS to the state-of-the-art method SConES, on a dataset of Arabidopsis Thaliana. SPADIS has better average phenotype prediction performance in 15 out of 17 phenotypes when same number of SNPs are selected, it identifies more candidate genes and runs faster. We investigate the use of Hi-C data to construct SNP-SNP network in the context of SNP selection problem for the first time and it yields consistent improvements in regression performance.

Three dimensional organization of the genome is emerging as an important determinant of cell-type specific expression and is implicated in many diseases, including cancer. Hi-C is a type of high-throughput chromosome conformation capture (3C) assay used to study three-dimensional organization of the genome. Analysis of Hi-C data has shown that the genome is organized into higher-order organizational units such as compartments and topologically associated domains (TADs). We present a non-negative matrix factorization approach, commonly used for clustering and dimensionality reduction, to infer clusters of regions from Hi-C data. To preserve the spatial dependency of Hi-C data (i.e. closer regions interact more with each other), we impose regularization on NMF with the nearest-neighbor graph of each genomic loci. Our results show that NMF and graph-regularized NMF are both important to discover clusters that exhibit a significant association with the presence of CTCF binding at the cluster boundaries and are robust to simulated sparsity and lower sequence depth.

Hi-C sequencing technology provides key insights into the 3D structures of the human genome. Although peak detection from Hi-C experiments is a well-studied problem, quantitative comparison of Hi-C (often referred as "differential (interaction) analysis" problem) across different cellular conditions largely depends on methods borrowed from RNA-seq data analysis. Such comparisons have critical shortcomings involving testing a large collection of hypotheses in large-scale Hi-C studies. As a result, these existing strategies for detecting differential interactions fail to control the rate of false discovery for reported findings in many simulations and experimental Hi-C studies, hindering their comparative analysis.

Here, we present TreeHiC, the first hierarchical testing procedure for quantitative comparison applied to Hi-C. We demonstrate that this framework can detect differential interactions while assuring control of the FDR in complex large-scale Hi-C studies under a wide range of settings. It also is considerably more powerful than existing methods, especially in sparse testing problems where number of hypotheses could be millions with a weak signal-to-noise ratio. Additionally, while the current version of TreeHiC implements methodology pertaining to Hi-C differential analysis, it is easily extendable for other similar data such as ChIA-PET and HiChIP. TreeHiC is open source, and can be downloaded from \url{https://github.com/duydnguyen/TreeHiC}.

The principles of complex genome organization within the cell nucleus are not well understood. This talk will introduce some of our recent work to reveal several different aspects of nuclear genome organization, including sequence determinants for long-range chromatin interactions, chromosomal spatial positioning in terms of nuclear compartmentalization, and evolutionary patterns of DNA replication timing across different species.

Polymer-based simulations and experimental studies indicate the existence of a spatial dependency between the adjacent DNA fibers involved in the formation of chromatin loops. However, the existing strategies for detecting differential chromatin interactions assume that the interacting segments are spatially independent from the other segments nearby. To resolve this issue, we developed a new computational method, FIND, which considers the local spatial dependency between interacting loci. FIND uses a spatial Poisson process to detect differential chromatin interactions that show a significant difference in their interaction frequency and the interaction frequency of their neighbors. Simulation and biological data analysis show that FIND outperforms the widely used count-based methods and has a better signal-to-noise ratio.

Broad domain promoters and super enhancers are regulatory elements that govern cell-specific functions and harbor disease-associated sequence variants. These elements are characterized by distinct epigenomic profiles, such as expanded deposition of histone marks H3K27ac for super enhancers and H3K4me3 for broad domains, however little is known about how they interact with each other and the rest of the genome in three-dimensional chromatin space. Using network theory methods, we studied chromatin interactions between broad domains and super enhancers in three ENCODE cell lines (K562, MCF7, GM12878) obtained via ChIA-PET, Hi-C, and Hi-CHIP assays. In these networks, broad domains and super enhancers interact more frequently with each other compared to their typical counterparts. Network measures and graphlets revealed distinct connectivity patterns associated with these regulatory elements that are robust across cell types and alternative assays. Machine learning models showed that these connectivity patterns could effectively discriminate broad domains from typical promoters and super enhancers from typical enhancers. Finally, targets of broad domains in these networks were enriched in disease-causing SNPs of cognate cell types. Taken together these results suggest a robust and unique organization of the chromatin around broad domains and super enhancers: loci critical for pathologies and cell-specific functions.

Chromatin interactions measured by the 3C-based family of next generation technologies are becoming increasingly important for measuring the physical basis for regulatory interactions between different classes of functional domains in the genome. Software is needed to streamline analyses of these data and integrate them with custom genome annotations, RNA-seq, and gene ontologies. We introduce a new R package compatible with Bioconductor--Hi-C Annotation and Graphics Ensemble (HiCAGE)--to perform these tasks with minimum effort. In addition, the package contains a shiny/R web app interface to provide ready access to its functions.

Originally, HiC data has been used to predict A/B compartments and sub-compartments. Genomic regions falling into the same compartments (or sub-compartment) are more likely to interact with each other compared to regions falling into different compartments. It has been reported that compartments (or sub-compartments) show distinct genomic and epigenomic features.

We present CompartmentExplorer, the first method that predicts genomic compartments from ChIA-PET data and sub-compartments from HiC and ChIA-PET data. In addition to traditional A/B compartment prediction method, CompartmentExplorer implements an efficient method based on graph-embedding followed by k-means clustering to predict sub-compartments from 3D genome data.

Using CompartmentExplorer, we show that combined ChIA-PET data from CTCF and RNAPII factors can be utilized to predict A/B compartments with comparable accuracy to HiC data (91% agreement). Further, we compared CompartmentExplorer sub-compartment predictions with 3 different methods based on hidden Markov models (HMMs), K-means clustering, spectral clustering. Using different validation techniques, including network centrality measures, and functional assessment by epigenome and transcriptome data, we show CompartmentExplorer outperforms other methods. For instance, CompartmentExploere shows significantly (p-value <2.2e-16) lower closeness centrality and higher betweenness centrality within each sub-compartment than other methods. CompartmentExpolorer provides the most distinct epigenetic and transcriptomic profiles to different sub-compartment.

HiChIP or PLAC-seq techniques combine chromatin immunoprecipitation (ChIP) with genome-wide chromosome conformation capture (Hi-C) methods to identify chromatin contacts associated with a protein or histone modification of interest. Employing Hi-C specific computational methods on HiChIP data to identify significant interactions exhibits low recall and/or low precision. Our proposed statistical method and overall pipeline, FitHiChIP, detects significant interactions by characterizing distance-dependent decay and assay-specific biases. FitHiChIP identifies 1D peaks from HiChIP data, supports external peaks, and normalizes peaks and non-peaks separately for their coverage. Results on published data from three cell lines and three primary immune cell types (Fang 2016, Mumbach 2016, 2017) show that FitHiChIP recovers high percentage of reported interactions from ChIA-PET and Hi-C data, while reducing the number of false negatives originally reported for these datasets (e.g. enhancer contacts from IL2RA promoter in T cells). For H3K27ac-based HiChIP experiments from naïve CD4 T cells, FitHiChIP identifies nearly 90k significant interactions from each biological replicate, majority of which are links between and within enhancers and promoters, and 77% (66%) of which are reproducible between replicates of the same (different) donor. FitHiChIP supports multiple input formats and multiprocessing, and processes very high resolution (e.g., 1kb) HiChIP data within hours.

Understanding the factors and mechanisms responsible for 3D genome organization is essential for gaining insights into gene regulation. CTCF is known to be a key regulator in chromosome folding. However, understanding the relationship between core units of 3D genome and CTCF still remains challenging as both altered loops and TADs vs. intact overall organization of genome upon CTCF loss are supported by experimental studies. Here we examine the hypothesis that effects of CTCF on 3D genome organization cannot be generalized, are site-specific and require a systematic one-by-one knock-out study. While this is experimentally demanding, computational analysis of Hi-C data is also challenged due to 2D nature of the data. We used the nC-SAC method to generate 3D ensembles of chromatin with and without CTCF to understand the role of CTCF in promoter-enhancer interactions. We found that CTCF binding alone is not sufficient for the chromosome organization. We identified structural hotspots that are highly conserved and enriched with CTCF binding that are likely responsible for bringing enhancers and promoters together. Our method provides a powerful tool for deciphering the structural units of chromatin that are enriched with important features like conservation or CTCF binding for assessing their impact on gene regulation.

The 3D conformation of DNA in the nucleus impacts gene expression, DNA replication, and multiple human diseases. The field of genome-wide chromosome conformation studies has advanced dramatically over the last decade, driven by advances in sequencing technology and development of novel assays such as Hi-C. In this talk, I will discuss some of the computational challenges posed by this type of large-scale data, including questions of statistical significance of observed contacts as well as inference of 3D structures. I will also describe how single-cell variants of the Hi-C assay can help us to understand cell-to-cell variability in 3D structure along developmental and cell cycle time scales

Nuclear organization of genomic DNA affects processes of DNA damage and repair, yet its effects on mutational landscapes in cancer genomes remain unclear. Here we analyzed genome-wide somatic mutations from 366 samples of six cancer types. We found that lamina-associated regions, which are typically localized at the nuclear periphery, displayed higher somatic mutation frequencies than did the interlamina regions at the nuclear core. This effect was observed even after adjustment for features such as GC percentage, chromatin, and replication timing. Furthermore, mutational signatures differed between the nuclear core
and periphery, thus indicating differences in the patterns of DNA-damage or DNA-repair processes. For instance, smoking and UV-related signatures, as well as substitutions at certain motifs, were more enriched in the nuclear periphery. Thus, the nuclear architecture may influence mutational landscapes in cancer genomes beyond the previously described effects of chromatin structure and replication timing.

Structural disorder of chromatin interactions (SDCI) is widely observed in chromosomes of cancer cells at different spatial scales and is one of the hallmarks of human cancers. The discovery of recurrent SDCIs and their molecular consequences for disrupting gene regulation and transcription is of great importance for not only understanding pathogenic mechanisms, but also providing diagnostic and prognostic information in the clinic. We provide a novel and unbiased method for comparing the interaction signals of ChIA-PET data. By analyzing PolII and CTCF ChIA-PET data of cancer cell lines, we successfully detected a large number of cancer-associated disorders of chromatin interactions that consequentially cause dysregulations of multiple oncogenes in various cancer types. Detailed analysis of these interactions revealed dramatic amplification-based regulation of oncogenes. Thus, our work provides disorders of chromatin interactions and their associated genes for systematically understanding the causality of gene dysregulation during cancer development.

To determine genetic factors, causing variation in longevity, several genome-wide association studies (GWAS) have been carried out on panels of long-lived individuals. Most studies tend to have little impact due to small sample sizes. For this reason model organisms such as Drosophila melanogaster have become increasingly important in identifying genetic factors affecting longevity.
In this study a network approach was used for predicting novel genes/genomic regions/single nucleotide polymorphisms (SNPs), playing a role in longevity, by integrating three-dimensional (3D) chromosome interaction data and two GWAS datasets (Burke et al. 2013; Ivanov et al. 2015). We hypothesise that 3D architecture of the Drosophila genome dictates the co-location of specific genes/genomic regions. Genes and/or SNPs, residing within these co-located genomic regions, may influence longevity either independently or have a cumulative effect on longevity. To identify influential nodes, the properties of networks were calculated (clustering, modularity and Page Rank). These nodes were further analysed using Gene Ontology.

References
Burke, M.K. et al. 2013. Genome-wide association study of extreme longevity in Drosophila melanogaster. Genome biology and evolution, 6:1-11.
Ivanov, D.K. et al. 2015. Longevity GWAS using the Drosophila genetic reference panel. Journals of Gerontology Series A: Biomedical Sciences and Medical Sciences, 70: 1470-1478.

Understanding the three-dimensional (3D) architecture of chromatin and its relation to gene expression and regulation is fundamental to understanding how the genome functions. Advances in Hi-C technology now permit us to study 3D genome organization, but we still lack an understanding of the structural dynamics of chromosomes. The dynamic couplings between regions separated by large genomic distances (>50 Mb) have yet to be characterized. We adapted a well-established protein-modeling framework, the Gaussian Network Model (GNM), to model chromatin dynamics using Hi-C data. We show that the GNM can identify spatial couplings at multiple scales: it can quantify the correlated fluctuations in the positions of gene loci, find large genomic compartments and smaller topologically-associating domains (TADs) that undergo en bloc movements, and identify dynamically coupled distal regions along the chromosomes. We show that the predictions of the GNM correlate well with genome-wide experimental measurements. We use the GNM to identify novel cross-correlated distal domains (CCDDs) representing pairs of regions distinguished by their long-range dynamic coupling and show that CCDDs are associated with increased gene co-expression. Together, these results show that GNM provides a mathematically well-founded unified framework for modeling chromatin dynamics and assessing the structural basis of genome-wide observations.

Current Hi-C analysis approaches are unable to account for reads that align to multiple locations, and hence underestimate biological signal from repetitive regions of genomes. We developed mHi-C, a multi-read mapping strategy to probabilistically allocate Hi-C multi-reads. mHi-C exhibited superior performance over utilizing only uni- reads and heuristic approaches aimed at rescuing multi-reads on benchmarks. Specifically, mHi-C increased the sequencing depth by an average of 20% leading to higher reproducibility of contact matrices and larger number of significant contacts across biological replicates. mHi-C also revealed biologically supported bona fide promoter-enhancer interactions and topologically associating domains involving repetitive genomic regions, thereby unlocking a previously masked portion of the genome for conformation capture studies.

Regulatory sequence elements such as enhancers can regulate the expression level of a gene hundreds of kilobases away through chromosomal looping, which brings regulatory elements in three-dimensional proximity to target genes. Although a large number of Chromosome Conformation Capture (3C) technologies have emerged, they have been limited to well-characterized model cell lines due to sequencing costs and the required number of cells for making reliable measurements at high resolution. We have developed a regression-based method, HiC-Reg, to generate in silico contact counts using one-dimensional regulatory signals. Our predicted counts are able to recapitulate significant interactions identified in true count data, are enriched for CTCF bi-directional motifs and can be used to identify topologically associated domains. Additionally, we leveraged chromatin mark data from the Roadmap Epigenomics Mapping Consortium to generate predictions in 22 human cell lines. To verify our predictions in a completely independent cell line, we compared our predicted counts in embryonic stem cells, which have low resolution Hi-C data and found good agreement in the true and predicted counts. Taken together our regression-based framework can generate high resolution profiles of contact counts that capture individual pair-level as well as higher-order organizational units of chromosome conformation.