Cell-type-specific prediction of 3D chromatin organization enables high-throughput in silico genetic screening

Machine and Deep Learning Methods for Predicting 3D Genome Organization

Three-dimensional (3D) chromatin interactions, such as enhancer-promoter interactions (EPIs), loops, topologically associating domains (TADs), and A/B compartments, play critical roles in a wide range of cellular processes by regulating gene expression. Recent development of chromatin conformation capture technologies has enabled genome-wide profiling of various 3D structures, even with single cells. However, current catalogs of 3D structures remain incomplete and unreliable due to differences in technology, tools, and low data resolution. Machine learning methods have emerged as an alternative to obtain missing 3D interactions and/or improve resolution. Such methods frequently use genome annotation data (ChIP-seq, DNAse-seq, etc.), DNA sequencing information (k-mers and transcription factor binding site (TFBS) motifs), and other genomic properties to learn the associations between genomic features and chromatin interactions. In this review, we discuss computational tools for predicting three types of 3D interactions (EPIs, chromatin interactions, and TAD boundaries) and analyze their pros and cons. We also point out obstacles to the computational prediction of 3D interactions and suggest future research directions.

Prediction of the 3D cancer genome from whole-genome sequencing using InfoHiC

The 3D genome prediction in cancer is crucial for uncovering the impact of structural variations (SVs) on tumorigenesis, especially when they are present in noncoding regions. We present InfoHiC, a systemic framework for predicting the 3D cancer genome directly from whole-genome sequencing (WGS). InfoHiC utilizes contig-specific copy number encoding on the SV contig assembly, and performs a contig-to-total Hi-C conversion for the cancer Hi-C prediction from multiple SV contigs. We showed that InfoHiC can predict 3D genome folding from all types of SVs using breast cancer cell line data. We applied it to WGS data of patients with breast cancer and pediatric patients with medulloblastoma, and identified neo topologically associating domains. For breast cancer, we discovered super-enhancer hijacking events associated with oncogenic overexpression and poor survival outcomes. For medulloblastoma, we found SVs in noncoding regions that caused super-enhancer hijacking events of medulloblastoma driver genes (GFI1, GFI1B, and PRDM6). In addition, we provide trained models for cancer Hi-C prediction from WGS at https://github.com/dmcb-gist/InfoHiC , uncovering the impacts of SVs in cancer patients and revealing novel therapeutic targets.

ImputeHiFI: An Imputation Method for Multiplexed DNA FISH Data by Utilizing Single-Cell Hi-C and RNA FISH Data

Although multiplexed DNA fluorescence in situ hybridization (FISH) enables tracking the spatial localization of thousands of genomic loci using probes within individual cells, the high rates of undetected probes impede the depiction of 3D chromosome structures. Current data imputation methods neither utilize single-cell Hi-C data, which elucidate 3D genome architectures using sequencing nor leverage multimodal RNA FISH data that reflect cell-type information, limiting the effectiveness of these methods in complex tissues such as the mouse brain. To this end, a novel multiplexed DNA FISH imputation method named ImputeHiFI is proposed, which fully utilizes the complementary structural information from single-cell Hi-C data and the cell type signature from RNA FISH data to obtain a high-fidelity and complete spatial location of chromatin loci. ImputeHiFI enhances cell clustering, compartment identification, and cell subtype detection at the single-cell level in the mouse brain. ImputeHiFI improves the recognition of cell-type-specific loops in three high-resolution datasets. In short, ImputeHiFI is a powerful tool capable of imputing multiplexed DNA FISH data from various resolutions and imaging protocols, facilitating studies of 3D genome structures and functions.

Systematic identification of interchromosomal interaction networks supports the existence of specialized RNA factories

Most studies of genome organization have focused on intrachromosomal (cis) contacts because they harbor key features such as DNA loops and topologically associating domains. Interchromosomal (trans) contacts have received much less attention, and tools for interrogating potential biologically relevant trans structures are lacking. Here, we develop a computational framework that uses Hi-C data to identify sets of loci that jointly interact in trans This method, trans-C, initiates probabilistic random walks with restarts from a set of seed loci to traverse an input Hi-C contact network, thereby identifying sets of trans-contacting loci. We validate trans-C in three increasingly complex models of established trans contacts: the Plasmodium falciparum var genes, the mouse olfactory receptor "Greek islands," and the human RBM20 cardiac splicing factory. We then apply trans-C to systematically test the hypothesis that genes coregulated by the same trans-acting element (i.e., a transcription or splicing factor) colocalize in three dimensions to form "RNA factories" that maximize the efficiency and accuracy of RNA biogenesis. We find that many loci with multiple binding sites of the same DNA-binding proteins interact with one another in trans, especially those bound by factors with intrinsically disordered domains. Similarly, clustered binding of a subset of RNA-binding proteins correlates with trans interaction of the encoding loci. We observe that these trans-interacting loci are close to nuclear speckles. These findings support the existence of trans- interacting chromatin domains (TIDs) driven by RNA biogenesis. Trans-C provides an efficient computational framework for studying these and other types of trans interactions, empowering studies of a poorly understood aspect of genome architecture.

An integrated view of the structure and function of the human 4D nucleome

The dynamic three-dimensional (3D) organization of the human genome (the "4D Nucleome") is closely linked to genome function. Here, we integrate a wide variety of genomic data generated by the 4D Nucleome Project to provide a detailed view of human 3D genome organization in widely used embryonic stem cells (H1-hESCs) and immortalized fibroblasts (HFFc6). We provide extensive benchmarking of 3D genome mapping assays and integrate these diverse datasets to annotate spatial genomic features across scales. The data reveal a rich complexity of chromatin domains and their sub-nuclear positions, and over one hundred thousand structural loops and promoter-enhancer interactions. We developed 3D models of population-based and individual cell-to-cell variation in genome structure, establishing connections between chromosome folding, nuclear organization, chromatin looping, gene transcription, and DNA replication. We demonstrate the use of computational methods to predict genome folding from DNA sequence, uncovering potential effects of genetic variants on genome structure and function. Together, this comprehensive analysis contributes insights into human genome organization and enhances our understanding of connections between the regulation of genome function and 3D genome organization in general.

HiCDiffusion - diffusion-enhanced, transformer-based prediction of chromatin interactions from DNA sequences

Prediction of chromatin interactions from DNA sequence has been a significant research challenge in the last couple of years. Several solutions have been proposed, most of which are based on encoder-decoder architecture, where 1D sequence is convoluted, encoded into the latent representation, and then decoded using 2D convolutions into the Hi-C pairwise chromatin spatial proximity matrix. Those methods, while obtaining high correlation scores and improved metrics, produce Hi-C matrices that are artificial - they are blurred due to the deep learning model architecture. In our study, we propose the HiCDiffusion, sequence-only model that addresses this problem. We first train the encoder-decoder neural network and then use it as a component of the diffusion model - where we guide the diffusion using a latent representation of the sequence, as well as the final output from the encoder-decoder. That way, we obtain the high-resolution Hi-C matrices that not only better resemble the experimental results - improving the Fréchet inception distance by an average of 11 times, with the highest improvement of 56 times - but also obtain similar classic metrics to current state-of-the-art encoder-decoder architectures used for the task.

Prediction and functional interpretation of inter-chromosomal genome architecture from DNA sequence with TwinC

Three-dimensional nuclear DNA architecture comprises well-studied intra-chromosomal (cis) folding and less characterized inter-chromosomal (trans) interfaces. Current predictive models of 3D genome folding can effectively infer pairwise cis-chromatin interactions from the primary DNA sequence but generally ignore trans contacts. There is an unmet need for robust models of trans-genome organization that provide insights into their underlying principles and functional relevance. We present TwinC, an interpretable convolutional neural network model that reliably predicts trans contacts measurable through genome-wide chromatin conformation capture (Hi-C). TwinC uses a paired sequence design from replicate Hi-C experiments to learn single base pair relevance in trans interactions across two stretches of DNA. The method achieves high predictive accuracy (AUROC=0.80) on a cross-chromosomal test set from Hi-C experiments in heart tissue. Mechanistically, the neural network learns the importance of compartments, chromatin accessibility, clustered transcription factor binding and G-quadruplexes in forming trans contacts. In summary, TwinC models and interprets trans genome architecture, shedding light on this poorly understood aspect of gene regulation.

Polymer Physics Models Reveal Structural Folding Features of Single-Molecule Gene Chromatin Conformations

Here, we employ polymer physics models of chromatin to investigate the 3D folding of a 2 Mb wide genomic region encompassing the human LTN1 gene, a crucial DNA locus involved in key cellular functions. Through extensive Molecular Dynamics simulations, we reconstruct in silico the ensemble of single-molecule LTN1 3D structures, which we benchmark against recent in situ Hi-C 2.0 data. The model-derived single molecules are then used to predict structural folding features at the single-cell level, providing testable predictions for super-resolution microscopy experiments.

Super-silencer perturbation by EZH2 and REST inhibition leads to large loss of chromatin interactions and reduction in cancer growth

Human silencers have been shown to regulate developmental gene expression. However, the functional importance of human silencers needs to be elucidated, such as whether they can form 'super-silencers' and whether they are linked to cancer progression. Here, we show two silencer components of the FGF18 gene can cooperate through compensatory chromatin interactions to form a super-silencer. Double knockout of two silencers exhibited synergistic upregulation of FGF18 expression and changes in cell identity. To perturb the super-silencers, we applied combinational treatment of an enhancer of zeste homolog 2 inhibitor GSK343, and a repressor element 1-silencing transcription factor inhibitor, X5050 ('GR'). Interestingly, GR led to severe loss of topologically associated domains and loops, which were associated with reduced CTCF and TOP2A mRNA levels. Moreover, GR synergistically upregulated super-silencer-controlled genes related to cell cycle, apoptosis and DNA damage, leading to anticancer effects in vivo. Overall, our data demonstrated a super-silencer example and showed that GR can disrupt super-silencers, potentially leading to cancer ablation.

Identifying genetic variants associated with chromatin looping and genome function

Here we present a comprehensive HiChIP dataset on naïve CD4 T cells (nCD4) from 30 donors and identify QTLs that associate with genotype-dependent and/or allele-specific variation of HiChIP contacts defining loops between active regulatory regions (iQTLs). We observe a substantial overlap between iQTLs and previously defined eQTLs and histone QTLs, and an enrichment for fine-mapped QTLs and GWAS variants. Furthermore, we describe a distinct subset of nCD4 iQTLs, for which the significant variation of chromatin contacts in nCD4 are translated into significant eQTL trends in CD4 T cell memory subsets. Finally, we define connectivity-QTLs as iQTLs that are significantly associated with concordant genotype-dependent changes in chromatin contacts over a broad genomic region (e.g., GWAS SNP in the RNASET2 locus). Our results demonstrate the importance of chromatin contacts as a complementary modality for QTL mapping and their power in identifying previously uncharacterized QTLs linked to cell-specific gene expression and connectivity.

Enhancer-promoter hubs organize transcriptional networks promoting oncogenesis and drug resistance

Recent advances in high-resolution mapping of spatial interactions among regulatory elements support the existence of complex topological assemblies of enhancers and promoters known as enhancer-promoter hubs or cliques. Yet, organization principles of these multi-interacting enhancer-promoter hubs and their potential role in regulating gene expression in cancer remain unclear. Here, we systematically identify enhancer-promoter hubs in breast cancer, lymphoma, and leukemia. We find that highly interacting enhancer-promoter hubs form at key oncogenes and lineage-associated transcription factors potentially promoting oncogenesis of these diverse cancer types. Genomic and optical mapping of interactions among enhancer and promoter elements further show that topological alterations in hubs coincide with transcriptional changes underlying acquired resistance to targeted therapy in T cell leukemia and B cell lymphoma. Together, our findings suggest that enhancer-promoter hubs are dynamic and heterogeneous topological assemblies with the potential to control gene expression circuits promoting oncogenesis and drug resistance.

EPInformer: a scalable deep learning framework for gene expression prediction by integrating promoter-enhancer sequences with multimodal epigenomic data

Transcriptional regulation, critical for cellular differentiation and adaptation to environmental changes, involves coordinated interactions among DNA sequences, regulatory proteins, and chromatin architecture. Despite extensive data from consortia like ENCODE, understanding the dynamics of cis-regulatory elements (CREs) in gene expression remains challenging. Deep learning is a powerful tool for learning gene expression and epigenomic signals from DNA sequences, exhibiting superior performance compared to conventional machine learning approaches. However, even the most advanced deep learning-based methods may fall short in capturing the regulatory effects of distal elements such as enhancers, limiting their predictive accuracy. In addition, these methods may require significant resources to train or to adapt to newly generated data. To address these challenges, we present EPInformer, a scalable deep-learning framework for predicting gene expression by integrating promoter-enhancer interactions with their sequences, epigenomic signals, and chromatin contacts. Our model outperforms existing gene expression prediction models in rigorous cross-chromosome validation, accurately recapitulates enhancer-gene interactions validated by CRISPR perturbation experiments, and identifies crucial transcription factor motifs within regulatory sequences. EPInformer is available as open-source software at https://github.com/pinellolab/EPInformer.

Modelling DNA damage-repair and beyond

The paper presents a review of mechanistic modelling studies of DNA damage and DNA repair, and consequences to follow in mammalian cell nucleus. We hypothesize DNA deletions are consequences of repair of double strand breaks leading to the modifications of genome that play crucial role in long term development of genetic inheritance and diseases. The aim of the paper is to review formation mechanisms underlying naturally occurring DNA deletions in the human genome and their potential relevance for bridging the gap between induced DNA double strand breaks and deletions in damaged human genome from endogenous and exogenous events. The model of the cell nucleus presented enables simulation of DNA damage at molecular level identifying the spectrum of damage induced in all chromosomal territories and loops. Our mechanistic modelling of DNA repair for double stand breaks (DSB), single strand breaks (SSB) and base damage (BD), shows the complexity of DNA damage is responsible for the longer repair times and the reason for the biphasic feature of mammalian cells repair curves. In the absence of experimentally determined data, the mechanistic model of repair predicts the in vivo rate constants for the proteins involved in the repair of DSB, SSB, and of BD.

Applying interpretable machine learning in computational biology-pitfalls, recommendations and opportunities for new developments

Recent advances in machine learning have enabled the development of next-generation predictive models for complex computational biology problems, thereby spurring the use of interpretable machine learning (IML) to unveil biological insights. However, guidelines for using IML in computational biology are generally underdeveloped. We provide an overview of IML methods and evaluation techniques and discuss common pitfalls encountered when applying IML methods to computational biology problems. We also highlight open questions, especially in the era of large language models, and call for collaboration between IML and computational biology researchers.

Deep Learning Sequence Models for Transcriptional Regulation

Deciphering the regulatory code of gene expression and interpreting the transcriptional effects of genome variation are critical challenges in human genetics. Modern experimental technologies have resulted in an abundance of data, enabling the development of sequence-based deep learning models that link patterns embedded in DNA to the biochemical and regulatory properties contributing to transcriptional regulation, including modeling epigenetic marks, 3D genome organization, and gene expression, with tissue and cell-type specificity. Such methods can predict the functional consequences of any noncoding variant in the human genome, even rare or never-before-observed variants, and systematically characterize their consequences beyond what is tractable from experiments or quantitative genetics studies alone. Recently, the development and application of interpretability approaches have led to the identification of key sequence patterns contributing to the predicted tasks, providing insights into the underlying biological mechanisms learned and revealing opportunities for improvement in future models.

Enhancer-promoter hubs organize transcriptional networks promoting oncogenesis and drug resistance

Recent advances in high-resolution mapping of spatial interactions among regulatory elements support the existence of complex topological assemblies of enhancers and promoters known as enhancer-promoter hubs or cliques. Yet, organization principles of these multi-interacting enhancer-promoter hubs and their potential role in regulating gene expression in cancer remains unclear. Here, we systematically identified enhancer-promoter hubs in breast cancer, lymphoma, and leukemia. We found that highly interacting enhancer-promoter hubs form at key oncogenes and lineage-associated transcription factors potentially promoting oncogenesis of these diverse cancer types. Genomic and optical mapping of interactions among enhancer and promoter elements further showed that topological alterations in hubs coincide with transcriptional changes underlying acquired resistance to targeted therapy in T cell leukemia and B cell lymphoma. Together, our findings suggest that enhancer-promoter hubs are dynamic and heterogeneous topological assemblies with the potential to control gene expression circuits promoting oncogenesis and drug resistance.

scGrapHiC: deep learning-based graph deconvolution for Hi-C using single cell gene expression

SUMMARY

Single-cell Hi-C (scHi-C) protocol helps identify cell-type-specific chromatin interactions and sheds light on cell differentiation and disease progression. Despite providing crucial insights, scHi-C data is often underutilized due to the high cost and the complexity of the experimental protocol. We present a deep learning framework, scGrapHiC, that predicts pseudo-bulk scHi-C contact maps using pseudo-bulk scRNA-seq data. Specifically, scGrapHiC performs graph deconvolution to extract genome-wide single-cell interactions from a bulk Hi-C contact map using scRNA-seq as a guiding signal. Our evaluations show that scGrapHiC, trained on seven cell-type co-assay datasets, outperforms typical sequence encoder approaches. For example, scGrapHiC achieves a substantial improvement of 23.2% in recovering cell-type-specific Topologically Associating Domains over the baselines. It also generalizes to unseen embryo and brain tissue samples. scGrapHiC is a novel method to generate cell-type-specific scHi-C contact maps using widely available genomic signals that enables the study of cell-type-specific chromatin interactions.

AVAILABILITY AND IMPLEMENTATION

The GitHub link: https://github.com/rsinghlab/scGrapHiC contains the source code of scGrapHiC and associated scripts to preprocess publicly available datasets to produce the results and visualizations we have discuss in this manuscript.

SuPreMo: a computational tool for streamlining in silico perturbation using sequence-based predictive models

SUMMARY

The increasing development of sequence-based machine learning models has raised the demand for manipulating sequences for this application. However, existing approaches to edit and evaluate genome sequences using models have limitations, such as incompatibility with structural variants, challenges in identifying responsible sequence perturbations, and the need for vcf file inputs and phased data. To address these bottlenecks, we present Sequence Mutator for Predictive Models (SuPreMo), a scalable and comprehensive tool for performing and supporting in silico mutagenesis experiments. We then demonstrate how pairs of reference and perturbed sequences can be used with machine learning models to prioritize pathogenic variants or discover new functional sequences.

AVAILABILITY AND IMPLEMENTATION

SuPreMo was written in Python, and can be run using only one line of code to generate both sequences and 3D genome disruption scores. The codebase, instructions for installation and use, and tutorials are on the GitHub page: https://github.com/ketringjoni/SuPreMo.

Hijacked enhancer-promoter and silencer-promoter loops in cancer

Recent work has shown that besides inducing fusion genes, structural variations (SVs) can also contribute to oncogenesis by disrupting the three-dimensional genome organization and dysregulating gene expression. At the chromatin-loop level, SVs can relocate enhancers or silencers from their original genomic loci to activate oncogenes or repress tumor suppressor genes. On a larger scale, different types of alterations in topologically associating domains (TADs) have been reported in cancer, such as TAD expansion, shuffling, and SV-induced neo-TADs. Furthermore, the transformation from normal cells to cancerous cells is usually coupled with active or repressive compartmental switches, and cancer-specific compartments have been proposed. This review discusses the sites, and the other latest advances in studying how SVs disrupt higher-order genome structure in cancer, which in turn leads to oncogene dysregulation. We also highlight the clinical implications of these changes and the challenges ahead in this field.

Single-cell mapping of cell-type specific chromatin architecture in the central nervous system

Determining how chromatin is structured in the nucleus is critical to studying its role in gene regulation. Recent advances in the analysis of single-cell chromatin architecture have considerably improved our understanding of cell-type-specific chromosome conformation and nuclear architecture. In this review, we discuss the methods used for analysis of 3D chromatin conformation, including sequencing-based methods, imaging-based techniques, and computational approaches. We further review the application of these methods in the study of the role of chromatin topology in neural development and disorders.

Learning Micro-C from Hi-C with diffusion models

In the last few years, Micro-C has shown itself as an improved alternative to Hi-C. It replaced the restriction enzymes in Hi-C assays with micrococcal nuclease (MNase), resulting in capturing nucleosome resolution chromatin interactions. The signal-to-noise improvement of Micro-C allows it to detect more chromatin loops than high-resolution Hi-C. However, compared with massive Hi-C datasets available in the literature, there are only a limited number of Micro-C datasets. To take full advantage of these Hi-C datasets, we present HiC2MicroC, a computational method learning and then predicting Micro-C from Hi-C based on the denoising diffusion probabilistic models (DDPM). We trained our DDPM and other regression models in human foreskin fibroblast (HFFc6) cell line and evaluated these methods in six different cell types at 5-kb and 1-kb resolution. Our evaluations demonstrate that both HiC2MicroC and regression methods can markedly improve Hi-C towards Micro-C, and our DDPM-based HiC2MicroC outperforms regression in various terms. First, HiC2MicroC successfully recovers most of the Micro-C loops even those not detected in Hi-C maps. Second, a majority of the HiC2MicroC-recovered loops anchor CTCF binding sites in a convergent orientation. Third, HiC2MicroC loops share genomic and epigenetic properties with Micro-C loops, including linking promoters and enhancers, and their anchors are enriched for structural proteins (CTCF and cohesin) and histone modifications. Lastly, we find our recovered loops are also consistent with the loops identified from promoter capture Micro-C (PCMicro-C) and Chromatin Interaction Analysis by Paired-End Tag Sequencing (ChIA-PET). Overall, HiC2MicroC is an effective tool for further studying Hi-C data with Micro-C as a template. HiC2MicroC is publicly available at https://github.com/zwang-bioinformatics/HiC2MicroC/.

Analysis of long-range chromatin contacts, compartments and looping between mouse embryonic stem cells, lens epithelium and lens fibers

BACKGROUND

Nuclear organization of interphase chromosomes involves individual chromosome territories, "open" and "closed" chromatin compartments, topologically associated domains (TADs) and chromatin loops. The DNA- and RNA-binding transcription factor CTCF together with the cohesin complex serve as major organizers of chromatin architecture. Cellular differentiation is driven by temporally and spatially coordinated gene expression that requires chromatin changes of individual loci of various complexities. Lens differentiation represents an advantageous system to probe transcriptional mechanisms underlying tissue-specific gene expression including high transcriptional outputs of individual crystallin genes until the mature lens fiber cells degrade their nuclei.

RESULTS

Chromatin organization between mouse embryonic stem (ES) cells, newborn (P0.5) lens epithelium and fiber cells were analyzed using Hi-C. Localization of CTCF in both lens chromatins was determined by ChIP-seq and compared with ES cells. Quantitative analyses show major differences between number and size of TADs and chromatin loop size between these three cell types. In depth analyses show similarities between lens samples exemplified by overlaps between compartments A and B. Lens epithelium-specific CTCF peaks are found in mostly methylated genomic regions while lens fiber-specific and shared peaks occur mostly within unmethylated DNA regions. Major differences in TADs and loops are illustrated at the ~ 500 kb Pax6 locus, encoding the critical lens regulatory transcription factor and within a larger ~ 15 Mb WAGR locus, containing Pax6 and other loci linked to human congenital diseases. Lens and ES cell Hi-C data (TADs and loops) together with ATAC-seq, CTCF, H3K27ac, H3K27me3 and ENCODE cis-regulatory sites are shown in detail for the Pax6, Sox1 and Hif1a loci, multiple crystallin genes and other important loci required for lens morphogenesis. The majority of crystallin loci are marked by unexpectedly high CTCF-binding across their transcribed regions.

CONCLUSIONS

Our study has generated the first data on 3-dimensional (3D) nuclear organization in lens epithelium and lens fibers and directly compared these data with ES cells. These findings generate novel insights into lens-specific transcriptional gene control, open new research avenues to study transcriptional condensates in lens fiber cells, and enable studies of non-coding genetic variants linked to cataract and other lens and ocular abnormalities.

Epigenomic insights into common human disease pathology

The epigenome-the chemical modifications and chromatin-related packaging of the genome-enables the same genetic template to be activated or repressed in different cellular settings. This multi-layered mechanism facilitates cell-type specific function by setting the local sequence and 3D interactive activity level. Gene transcription is further modulated through the interplay with transcription factors and co-regulators. The human body requires this epigenomic apparatus to be precisely installed throughout development and then adequately maintained during the lifespan. The causal role of the epigenome in human pathology, beyond imprinting disorders and specific tumour suppressor genes, was further brought into the spotlight by large-scale sequencing projects identifying that mutations in epigenomic machinery genes could be critical drivers in both cancer and developmental disorders. Abrogation of this cellular mechanism is providing new molecular insights into pathogenesis. However, deciphering the full breadth and implications of these epigenomic changes remains challenging. Knowledge is accruing regarding disease mechanisms and clinical biomarkers, through pathogenically relevant and surrogate tissue analyses, respectively. Advances include consortia generated cell-type specific reference epigenomes, high-throughput DNA methylome association studies, as well as insights into ageing-related diseases from biological 'clocks' constructed by machine learning algorithms. Also, 3rd-generation sequencing is beginning to disentangle the complexity of genetic and DNA modification haplotypes. Cell-free DNA methylation as a cancer biomarker has clear clinical utility and further potential to assess organ damage across many disorders. Finally, molecular understanding of disease aetiology brings with it the opportunity for exact therapeutic alteration of the epigenome through CRISPR-activation or inhibition.

Machine and deep learning methods for predicting 3D genome organization

Three-Dimensional (3D) chromatin interactions, such as enhancer-promoter interactions (EPIs), loops, Topologically Associating Domains (TADs), and A/B compartments play critical roles in a wide range of cellular processes by regulating gene expression. Recent development of chromatin conformation capture technologies has enabled genome-wide profiling of various 3D structures, even with single cells. However, current catalogs of 3D structures remain incomplete and unreliable due to differences in technology, tools, and low data resolution. Machine learning methods have emerged as an alternative to obtain missing 3D interactions and/or improve resolution. Such methods frequently use genome annotation data (ChIP-seq, DNAse-seq, etc.), DNA sequencing information (k-mers, Transcription Factor Binding Site (TFBS) motifs), and other genomic properties to learn the associations between genomic features and chromatin interactions. In this review, we discuss computational tools for predicting three types of 3D interactions (EPIs, chromatin interactions, TAD boundaries) and analyze their pros and cons. We also point out obstacles of computational prediction of 3D interactions and suggest future research directions.

Computational methods for analysing multiscale 3D genome organization

Recent progress in whole-genome mapping and imaging technologies has enabled the characterization of the spatial organization and folding of the genome in the nucleus. In parallel, advanced computational methods have been developed to leverage these mapping data to reveal multiscale three-dimensional (3D) genome features and to provide a more complete view of genome structure and its connections to genome functions such as transcription. Here, we discuss how recently developed computational tools, including machine-learning-based methods and integrative structure-modelling frameworks, have led to a systematic, multiscale delineation of the connections among different scales of 3D genome organization, genomic and epigenomic features, functional nuclear components and genome function. However, approaches that more comprehensively integrate a wide variety of genomic and imaging datasets are still needed to uncover the functional role of 3D genome structure in defining cellular phenotypes in health and disease.

Epigenetic regulatory layers in the 3D nucleus

Nearly 7 decades have elapsed since Francis Crick introduced the central dogma of molecular biology, as part of his ideas on protein synthesis, setting the fundamental rules of sequence information transfer from DNA to RNAs and proteins. We have since learned that gene expression is finely tuned in time and space, due to the activities of RNAs and proteins on regulatory DNA elements, and through cell-type-specific three-dimensional conformations of the genome. Here, we review major advances in genome biology and discuss a set of ideas on gene regulation and highlight how various biomolecular assemblies lead to the formation of structural and regulatory features within the nucleus, with roles in transcriptional control. We conclude by suggesting further developments that will help capture the complex, dynamic, and often spatially restricted events that govern gene expression in mammalian cells.

ChIPr: accurate prediction of cohesin-mediated 3D genome organization from 2D chromatin features

The three-dimensional genome organization influences diverse nuclear processes. Here we present Chromatin Interaction Predictor (ChIPr), a suite of regression models based on deep neural networks, random forest, and gradient boosting to predict cohesin-mediated chromatin interaction strength between any two loci in the genome. The predictions of ChIPr correlate well with ChIA-PET data in four cell lines. The standard ChIPr model requires three experimental inputs: ChIP-Seq signals for RAD21, H3K27ac, and H3K27me3 but works well with just RAD21 signal. Integrative analysis reveals novel insights into the role of CTCF motif, its orientation, and CTCF binding on cohesin-mediated chromatin interactions.

SuPreMo: a computational tool for streamlining in silico perturbation using sequence-based predictive models

Computationally editing genome sequences is a common bioinformatics task, but current approaches have limitations, such as incompatibility with structural variants, challenges in identifying responsible sequence perturbations, and the need for vcf file inputs and phased data. To address these bottlenecks, we present Sequence Mutator for Predictive Models (SuPreMo), a scalable and comprehensive tool for performing in silico mutagenesis. We then demonstrate how pairs of reference and perturbed sequences can be used with machine learning models to prioritize pathogenic variants or discover new functional sequences.

Predictive analyses of regulatory sequences with EUGENe

Deep learning has become a popular tool to study cis-regulatory function. Yet efforts to design software for deep-learning analyses in regulatory genomics that are findable, accessible, interoperable and reusable (FAIR) have fallen short of fully meeting these criteria. Here we present elucidating the utility of genomic elements with neural nets (EUGENe), a FAIR toolkit for the analysis of genomic sequences with deep learning. EUGENe consists of a set of modules and subpackages for executing the key functionality of a genomics deep learning workflow: (1) extracting, transforming and loading sequence data from many common file formats; (2) instantiating, initializing and training diverse model architectures; and (3) evaluating and interpreting model behavior. We designed EUGENe as a simple, flexible and extensible interface for streamlining and customizing end-to-end deep-learning sequence analyses, and illustrate these principles through application of the toolkit to three predictive modeling tasks. We hope that EUGENe represents a springboard towards a collaborative ecosystem for deep-learning applications in genomics research.

Gene regulatory network inference in the era of single-cell multi-omics

The interplay between chromatin, transcription factors and genes generates complex regulatory circuits that can be represented as gene regulatory networks (GRNs). The study of GRNs is useful to understand how cellular identity is established, maintained and disrupted in disease. GRNs can be inferred from experimental data - historically, bulk omics data - and/or from the literature. The advent of single-cell multi-omics technologies has led to the development of novel computational methods that leverage genomic, transcriptomic and chromatin accessibility information to infer GRNs at an unprecedented resolution. Here, we review the key principles of inferring GRNs that encompass transcription factor-gene interactions from transcriptomics and chromatin accessibility data. We focus on the comparison and classification of methods that use single-cell multimodal data. We highlight challenges in GRN inference, in particular with respect to benchmarking, and potential further developments using additional data modalities.

Sequence-based machine learning reveals 3D genome differences between bonobos and chimpanzees

Phenotypic divergence between closely related species, including bonobos and chimpanzees (genus Pan), is largely driven by variation in gene regulation. The 3D structure of the genome mediates gene expression; however, genome folding differences in Pan are not well understood. Here, we apply machine learning to predict genome-wide 3D genome contact maps from DNA sequence for 56 bonobos and chimpanzees, encompassing all five extant lineages. We use a pairwise approach to estimate 3D divergence between individuals from the resulting contact maps in 4,420 1 Mb genomic windows. While most pairs were similar, ∼17% were predicted to be substantially divergent in genome folding. The most dissimilar maps were largely driven by single individuals with rare variants that produce unique 3D genome folding in a region. We also identified 89 genomic windows where bonobo and chimpanzee contact maps substantially diverged, including several windows harboring genes associated with traits implicated in Pan phenotypic divergence. We used in silico mutagenesis to identify 51 3D-modifying variants in these bonobo-chimpanzee divergent windows, finding that 34 or 66.67% induce genome folding changes via CTCF binding motif disruption. Our results reveal 3D genome variation at the population-level and identify genomic regions where changes in 3D folding may contribute to phenotypic differences in our closest living relatives.

In silico discovery of repetitive elements as key sequence determinants of 3D genome folding

Natural and experimental genetic variants can modify DNA loops and insulating boundaries to tune transcription, but it is unknown how sequence perturbations affect chromatin organization genome wide. We developed a deep-learning strategy to quantify the effect of any insertion, deletion, or substitution on chromatin contacts and systematically scored millions of synthetic variants. While most genetic manipulations have little impact, regions with CTCF motifs and active transcription are highly sensitive, as expected. Our unbiased screen and subsequent targeted experiments also point to noncoding RNA genes and several families of repetitive elements as CTCF-motif-free DNA sequences with particularly large effects on nearby chromatin interactions, sometimes exceeding the effects of CTCF sites and explaining interactions that lack CTCF. We anticipate that our disruption tracks may be of broad interest and utility as a measure of 3D genome sensitivity, and our computational strategies may serve as a template for biological inquiry with deep learning.

Predicting patient-specific enhancer-promoter interactions

Computational methods that can predict hard-to-measure modalities from those that are easier to measure, in a patient-specific manner, play a critical role in personalized medicine. In this issue of Cell Reports Methods, Khurana et al. present differential gene targets of accessible chromatin (DGTAC), an approach which predicts patient-specific enhancer-promoter interactions.

Multimodal learning of noncoding variant effects using genome sequence and chromatin structure

MOTIVATION

A growing amount of noncoding genetic variants, including single-nucleotide polymorphisms, are found to be associated with complex human traits and diseases. Their mechanistic interpretation is relatively limited and can use the help from computational prediction of their effects on epigenetic profiles. However, current models often focus on local, 1D genome sequence determinants and disregard global, 3D chromatin structure that critically affects epigenetic events.

RESULTS

We find that noncoding variants of unexpected high similarity in epigenetic profiles, with regards to their relatively low similarity in local sequences, can be largely attributed to their proximity in chromatin structure. Accordingly, we have developed a multimodal deep learning scheme that incorporates both data of 1D genome sequence and 3D chromatin structure for predicting noncoding variant effects. Specifically, we have integrated convolutional and recurrent neural networks for sequence embedding and graph neural networks for structure embedding despite the resolution gap between the two types of data, while utilizing recent DNA language models. Numerical results show that our models outperform competing sequence-only models in predicting epigenetic profiles and their use of long-range interactions complement sequence-only models in extracting regulatory motifs. They prove to be excellent predictors for noncoding variant effects in gene expression and pathogenicity, whether in unsupervised "zero-shot" learning or supervised "few-shot" learning.

AVAILABILITY AND IMPLEMENTATION

Codes and data can be accessed at https://github.com/Shen-Lab/ncVarPred-1D3D and https://zenodo.org/record/7975777.

ChromaFold predicts the 3D contact map from single-cell chromatin accessibility

The identification of cell-type-specific 3D chromatin interactions between regulatory elements can help to decipher gene regulation and to interpret the function of disease-associated non-coding variants. However, current chromosome conformation capture (3C) technologies are unable to resolve interactions at this resolution when only small numbers of cells are available as input. We therefore present ChromaFold, a deep learning model that predicts 3D contact maps and regulatory interactions from single-cell ATAC sequencing (scATAC-seq) data alone. ChromaFold uses pseudobulk chromatin accessibility, co-accessibility profiles across metacells, and predicted CTCF motif tracks as input features and employs a lightweight architecture to enable training on standard GPUs. Once trained on paired scATAC-seq and Hi-C data in human cell lines and tissues, ChromaFold can accurately predict both the 3D contact map and peak-level interactions across diverse human and mouse test cell types. In benchmarking against a recent deep learning method that uses bulk ATAC-seq, DNA sequence, and CTCF ChIP-seq to make cell-type-specific predictions, ChromaFold yields superior prediction performance when including CTCF ChIP-seq data as an input and comparable performance without. Finally, fine-tuning ChromaFold on paired scATAC-seq and Hi-C in a complex tissue enables deconvolution of chromatin interactions across cell subpopulations. ChromaFold thus achieves state-of-the-art prediction of 3D contact maps and regulatory interactions using scATAC-seq alone as input data, enabling accurate inference of cell-type-specific interactions in settings where 3C-based assays are infeasible.

Systematic mapping and modeling of 3D enhancer-promoter interactions in early mouse embryonic lineages reveal regulatory principles that determine the levels and cell-type specificity of gene expression

Mammalian embryogenesis commences with two pivotal and binary cell fate decisions that give rise to three essential lineages, the trophectoderm (TE), the epiblast (EPI) and the primitive endoderm (PrE). Although key signaling pathways and transcription factors that control these early embryonic decisions have been identified, the non-coding regulatory elements via which transcriptional regulators enact these fates remain understudied. To address this gap, we have characterized, at a genome-wide scale, enhancer activity and 3D connectivity in embryo-derived stem cell lines that represent each of the early developmental fates. We observed extensive enhancer remodeling and fine-scale 3D chromatin rewiring among the three lineages, which strongly associate with transcriptional changes, although there are distinct groups of genes that are irresponsive to topological changes. In each lineage, a high degree of connectivity or "hubness" positively correlates with levels of gene expression and enriches for cell-type specific and essential genes. Genes within 3D hubs also show a significantly stronger probability of coregulation across lineages, compared to genes in linear proximity or within the same contact domains. By incorporating 3D chromatin features, we build a novel predictive model for transcriptional regulation (3D-HiChAT), which outperformed models that use only 1D promoter or proximal variables in predicting levels and cell-type specificity of gene expression. Using 3D-HiChAT, we performed genome-wide in silico perturbations to nominate candidate functional enhancers and hubs in each cell lineage, and with CRISPRi experiments we validated several novel enhancers that control expression of one or more genes in their respective lineages. Our study comprehensively identifies 3D regulatory hubs associated with the earliest mammalian lineages and describes their relationship to gene expression and cell identity, providing a framework to understand lineage-specific transcriptional behaviors.

Probabilistic establishment of speckle-associated inter-chromosomal interactions

Inter-chromosomal interactions play a crucial role in genome organization, yet the organizational principles remain elusive. Here, we introduce a novel computational method to systematically characterize inter-chromosomal interactions using in situ Hi-C results from various cell types. Our method successfully identifies two apparently hub-like inter-chromosomal contacts associated with nuclear speckles and nucleoli, respectively. Interestingly, we discover that nuclear speckle-associated inter-chromosomal interactions are highly cell-type invariant with a marked enrichment of cell-type common super-enhancers (CSEs). Validation using DNA Oligopaint fluorescence in situ hybridization (FISH) shows a strong but probabilistic interaction behavior between nuclear speckles and CSE-harboring genomic regions. Strikingly, we find that the likelihood of speckle-CSE associations can accurately predict two experimentally measured inter-chromosomal contacts from Hi-C and Oligopaint DNA FISH. Our probabilistic establishment model well describes the hub-like structure observed at the population level as a cumulative effect of summing individual stochastic chromatin-speckle interactions. Lastly, we observe that CSEs are highly co-occupied by MAZ binding and MAZ depletion leads to significant disorganization of speckle-associated inter-chromosomal contacts. Taken together, our results propose a simple organizational principle of inter-chromosomal interactions mediated by MAZ-occupied CSEs.

Considerations and caveats for analyzing chromatin compartments

Genomes are organized into nuclear compartments, separating active from inactive chromatin. Chromatin compartments are readily visible in a large number of species by experiments that map chromatin conformation genome-wide. When analyzing these maps, a common step is the identification of genomic intervals that interact within A (active) and B (inactive) compartments. It has also become increasingly common to identify and analyze subcompartments. We review different strategies to identify A/B and subcompartment intervals, including a discussion of various machine-learning approaches to predict these features. We then discuss the strengths and limitations of current strategies and examine how these aspects of analysis may have impacted our understanding of chromatin compartments.

Comparing chromatin contact maps at scale: methods and insights

Comparing chromatin contact maps is an essential step in quantifying how three-dimensional (3D) genome organization shapes development, evolution, and disease. However, no gold standard exists for comparing contact maps, and even simple methods often disagree. In this study, we propose novel comparison methods and evaluate them alongside existing approaches using genome-wide Hi-C data and 22,500 in silico predicted contact maps. We also quantify the robustness of methods to common sources of biological and technical variation, such as boundary size and noise. We find that simple difference-based methods such as mean squared error are suitable for initial screening, but biologically informed methods are necessary to identify why maps diverge and propose specific functional hypotheses. We provide a reference guide, codebase, and benchmark for rapidly comparing chromatin contact maps at scale to enable biological insights into the 3D organization of the genome.