Predicting the human epigenome from DNA motifs

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.

TFscope: systematic analysis of the sequence features involved in the binding preferences of transcription factors

Characterizing the binding preferences of transcription factors (TFs) in different cell types and conditions is key to understand how they orchestrate gene expression. Here, we develop TFscope, a machine learning approach that identifies sequence features explaining the binding differences observed between two ChIP-seq experiments targeting either the same TF in two conditions or two TFs with similar motifs (paralogous TFs). TFscope systematically investigates differences in the core motif, nucleotide environment and co-factor motifs, and provides the contribution of each key feature in the two experiments. TFscope was applied to > 305 ChIP-seq pairs, and several examples are discussed.

Quiescence enables unrestricted cell fate in naive embryonic stem cells

Quiescence in stem cells is traditionally considered as a state of inactive dormancy or with poised potential. Naive mouse embryonic stem cells (ESCs) can enter quiescence spontaneously or upon inhibition of MYC or fatty acid oxidation, mimicking embryonic diapause in vivo. The molecular underpinning and developmental potential of quiescent ESCs (qESCs) are relatively unexplored. Here we show that qESCs possess an expanded or unrestricted cell fate, capable of generating both embryonic and extraembryonic cell types (e.g., trophoblast stem cells). These cells have a divergent metabolic landscape comparing to the cycling ESCs, with a notable decrease of the one-carbon metabolite S-adenosylmethionine. The metabolic changes are accompanied by a global reduction of H3K27me3, an increase of chromatin accessibility, as well as the de-repression of endogenous retrovirus MERVL and trophoblast master regulators. Depletion of methionine adenosyltransferase Mat2a or deletion of Eed in the polycomb repressive complex 2 results in removal of the developmental constraints towards the extraembryonic lineages. Our findings suggest that quiescent ESCs are not dormant but rather undergo an active transition towards an unrestricted cell fate.

Bottom-up data integration in polymer models of chromatin organization

Cellular functions crucially depend on the precise execution of complex biochemical reactions taking place on the chromatin fiber in the tightly packed environment of the cell nucleus. Despite the availability of large datasets probing this process from multiple angles, bottom-up frameworks that allow the incorporation of the sequence-specific nature of biochemistry in a unified model of 3D chromatin structure remain scarce. Here, we propose Sequence-Enhanced Magnetic Polymer (SEMPER), a novel stochastic polymer model that naturally incorporates observational data about sequence-driven biochemical processes, such as binding of transcription factor proteins, in a 3D model of chromatin structure. We introduce a novel approximate Bayesian algorithm to quantify a posteriori the relative importance of various factors, including the polymeric nature of DNA, in determining chromatin epigenetic state, thus providing a transparent way to generate biological hypotheses. Although accurate prediction of contact frequencies (a problem already extensively studied in the literature) is not our main aim, as a by-product of the inference procedure and without additional input from the genome 3D structure, our model can predict with reasonable accuracy some notable and nontrivial conformational features of chromatin folding within the nucleus. Our work highlights the importance of introducing physically realistic statistical models for predicting chromatin states from epigenetic data and opens the way to a new class of more systematic approaches to interpreting epigenomic data.

Widespread effects of DNA methylation and intra-motif dependencies revealed by novel transcription factor binding models

Several studies suggested that transcription factor (TF) binding to DNA may be impaired or enhanced by DNA methylation. We present MeDeMo, a toolbox for TF motif analysis that combines information about DNA methylation with models capturing intra-motif dependencies. In a large-scale study using ChIP-seq data for 335 TFs, we identify novel TFs that show a binding behaviour associated with DNA methylation. Overall, we find that the presence of CpG methylation decreases the likelihood of binding for the majority of methylation-associated TFs. For a considerable subset of TFs, we show that intra-motif dependencies are pivotal for accurately modelling the impact of DNA methylation on TF binding. We illustrate that the novel methylation-aware TF binding models allow to predict differential ChIP-seq peaks and improve the genome-wide analysis of TF binding. Our work indicates that simplistic models that neglect the effect of DNA methylation on DNA binding may lead to systematic underperformance for methylation-associated TFs.

Predicting regional somatic mutation rates using DNA motifs

How the locus-specificity of epigenetic modifications is regulated remains an unanswered question. A contributing mechanism is that epigenetic enzymes are recruited to specific loci by DNA binding factors recognizing particular sequence motifs (referred to as epi-motifs). Using these motifs to predict biological outputs depending on local epigenetic state such as somatic mutation rates would confirm their functionality. Here, we used DNA motifs including known TF motifs and epi-motifs as a surrogate of epigenetic signals to predict somatic mutation rates in 13 cancers at an average 23kbp resolution. We implemented an interpretable neural network model, called contextual regression, to successfully learn the universal relationship between mutations and DNA motifs, and uncovered motifs that are most impactful on the regional mutation rates such as TP53 and epi-motifs associated with H3K9me3. Furthermore, we identified genomic regions with significantly higher mutation rates than the expected values in each individual tumor and demonstrated that such cancer-related regions can accurately predict cancer types. Interestingly, we found that the same mutation signatures often have different contributions to cancer-related and cancer-independent regions, and we also identified the motifs with the most contribution to each mutation signature.

A single amino acid change led to structural and functional differentiation of PvHd1 to control flowering in switchgrass

Switchgrass, a forage and bioenergy crop, occurs as two main ecotypes with different but overlapping ranges of adaptation. The two ecotypes differ in a range of characteristics, including flowering time. Flowering time determines the duration of vegetative development and therefore biomass accumulation, a key trait in bioenergy crops. No causal variants for flowering time differences between switchgrass ecotypes have, as yet, been identified. In this study, we mapped a robust flowering time quantitative trait locus (QTL) on chromosome 4K in a biparental F2 population and characterized the flowering-associated transcription factor gene PvHd1, an ortholog of CONSTANS in Arabidopsis and Heading date 1 in rice, as the underlying causal gene. Protein modeling predicted that a serine to glycine substitution at position 35 (p.S35G) in B-Box domain 1 greatly altered the global structure of the PvHd1 protein. The predicted variation in protein compactness was supported in vitro by a 4 °C shift in denaturation temperature. Overexpressing the PvHd1-p.35S allele in a late-flowering CONSTANS-null Arabidopsis mutant rescued earlier flowering, whereas PvHd1-p.35G had a reduced ability to promote flowering, demonstrating that the structural variation led to functional divergence. Our findings provide us with a tool to manipulate the timing of floral transition in switchgrass cultivars and, potentially, expand their cultivation range.

Underlying causes for prevalent false positives and false negatives in STARR-seq data

Self-transcribing active regulatory region sequencing (STARR-seq) and its variants have been widely used to characterize enhancers. However, it has been reported that up to 87% of STARR-seq peaks are located in repressive chromatin and are not functional in the tested cells. While some of the STARR-seq peaks in repressive chromatin might be active in other cell/tissue types, some others might be false positives. Meanwhile, many active enhancers may not be identified by the current STARR-seq methods. Although methods have been proposed to mitigate systematic errors caused by the use of plasmid vectors, the artifacts due to the intrinsic limitations of current STARR-seq methods are still prevalent and the underlying causes are not fully understood. Based on predicted cis-regulatory modules (CRMs) and non-CRMs in the human genome as well as predicted active CRMs and non-active CRMs in a few human cell lines/tissues with STARR-seq data available, we reveal prevalent false positives and false negatives in STARR-seq peaks generated by major variants of STARR-seq methods and possible underlying causes. Our results will help design strategies to improve STARR-seq methods and interpret the results.

iCpG-Pos: an accurate computational approach for identification of CpG sites using positional features on single-cell whole genome sequence data

MOTIVATION

The investigation of DNA methylation can shed light on the processes underlying human well-being and help determine overall human health. However, insufficient coverage makes it challenging to implement single-stranded DNA methylation sequencing technologies, highlighting the need for an efficient prediction model. Models are required to create an understanding of the underlying biological systems and to project single-cell (methylated) data accurately.

RESULTS

In this study, we developed positional features for predicting CpG sites. Positional characteristics of the sequence are derived using data from CpG regions and the separation between nearby CpG sites. Multiple optimized classifiers and different ensemble learning approaches are evaluated. The OPTUNA framework is used to optimize the algorithms. The CatBoost algorithm followed by the stacking algorithm outperformed existing DNA methylation identifiers.

AVAILABILITY AND IMPLEMENTATION

The data and methodologies used in this study are openly accessible to the research community. Researchers can access the positional features and algorithms used for predicting CpG site methylation patterns. To achieve superior performance, we employed the CatBoost algorithm followed by the stacking algorithm, which outperformed existing DNA methylation identifiers. The proposed iCpG-Pos approach utilizes only positional features, resulting in a substantial reduction in computational complexity compared to other known approaches for detecting CpG site methylation patterns. In conclusion, our study introduces a novel approach, iCpG-Pos, for predicting CpG site methylation patterns. By focusing on positional features, our model offers both accuracy and efficiency, making it a promising tool for advancing DNA methylation research and its applications in human health and well-being.

Prioritizing genes associated with brain disorders by leveraging enhancer-promoter interactions in diverse neural cells and tissues

BACKGROUND

Prioritizing genes that underlie complex brain disorders poses a considerable challenge. Despite previous studies have found that they shared symptoms and heterogeneity, it remained difficult to systematically identify the risk genes associated with them.

METHODS

By using the CAGE (Cap Analysis of Gene Expression) read alignment files for 439 human cell and tissue types (including primary cells, tissues and cell lines) from FANTOM5 project, we predicted enhancer-promoter interactions (EPIs) of 439 cell and tissue types in human, and examined their reliability. Then we evaluated the genetic heritability of 17 diverse brain disorders and behavioral-cognitive phenotypes in each neural cell type, brain region, and developmental stage. Furthermore, we prioritized genes associated with brain disorders and phenotypes by leveraging the EPIs in each neural cell and tissue type, and analyzed their pleiotropy and functionality for different categories of disorders and phenotypes. Finally, we characterized the spatiotemporal expression dynamics of these associated genes in cells and tissues.

RESULTS

We found that identified EPIs showed activity specificity and network aggregation in cell and tissue types, and enriched TF binding in neural cells played key roles in synaptic plasticity and nerve cell development, i.e., EGR1 and SOX family. We also discovered that most neurological disorders exhibit heritability enrichment in neural stem cells and astrocytes, while psychiatric disorders and behavioral-cognitive phenotypes exhibit enrichment in neurons. Furthermore, our identified genes recapitulated well-known risk genes, which exhibited widespread pleiotropy between psychiatric disorders and behavioral-cognitive phenotypes (i.e., FOXP2), and indicated expression specificity in neural cell types, brain regions, and developmental stages associated with disorders and phenotypes. Importantly, we showed the potential associations of brain disorders with brain regions and developmental stages that have not been well studied.

CONCLUSIONS

Overall, our study characterized the gene-enhancer regulatory networks and genetic mechanisms in the human neural cells and tissues, and illustrated the value of reanalysis of publicly available genomic datasets.

Integrative modeling of lncRNA-chromatin interaction maps reveals diverse mechanisms of nuclear retention

BACKGROUND

Many long non-coding RNAs, known to be involved in transcriptional regulation, are enriched in the nucleus and interact with chromatin. However, their mechanisms of chromatin interaction and the served cellular functions are poorly understood. We sought to characterize the mechanisms of lncRNA nuclear retention by systematically mapping the sequence and chromatin features that distinguish lncRNA-interacting genomic segments.

RESULTS

We found DNA 5-mer frequencies to be predictive of chromatin interactions for all lncRNAs, suggesting sequence-specificity as a global theme in the interactome. Sequence features representing protein-DNA and protein-RNA binding motifs revealed potential mechanisms for specific lncRNAs. Complementary to these global themes, transcription-related features and DNA-RNA triplex formation potential were noted to be highly predictive for two mutually exclusive sets of lncRNAs. DNA methylation was also noted to be a significant predictor, but only when combined with other epigenomic features.

CONCLUSIONS

Taken together, our statistical findings suggest that a group of lncRNAs interacts with transcriptionally inactive chromatin through triplex formation, whereas another group interacts with transcriptionally active regions and is involved in DNA Damage Response (DDR) through formation of R-loops. Curiously, we observed a strong pattern of enrichment of 5-mers in four potentially interacting entities: lncRNA-bound DNA tiles, lncRNAs, miRNA seed sequences, and repeat elements. This finding points to a broad sequence-based network of interactions that may underlie regulation of fundamental cellular functions. Overall, this study reveals diverse sequence and chromatin features related to lncRNA-chromatin interactions, suggesting potential mechanisms of nuclear retention and regulatory function.

Recall DNA methylation levels at low coverage sites using a CNN model in WGBS

DNA methylation is an important regulator of gene transcription. WGBS is the gold-standard approach for base-pair resolution quantitative of DNA methylation. It requires high sequencing depth. Many CpG sites with insufficient coverage in the WGBS data, resulting in inaccurate DNA methylation levels of individual sites. Many state-of-arts computation methods were proposed to predict the missing value. However, many methods required either other omics datasets or other cross-sample data. And most of them only predicted the state of DNA methylation. In this study, we proposed the RcWGBS, which can impute the missing (or low coverage) values from the DNA methylation levels on the adjacent sides. Deep learning techniques were employed for the accurate prediction. The WGBS datasets of H1-hESC and GM12878 were down-sampled. The average difference between the DNA methylation level at 12× depth predicted by RcWGBS and that at >50× depth in the H1-hESC and GM2878 cells are less than 0.03 and 0.01, respectively. RcWGBS performed better than METHimpute even though the sequencing depth was as low as 12×. Our work would help to process methylation data of low sequencing depth. It is beneficial for researchers to save sequencing costs and improve data utilization through computational methods.

DeepFormer: a hybrid network based on convolutional neural network and flow-attention mechanism for identifying the function of DNA sequences

Identifying the function of DNA sequences accurately is an essential and challenging task in the genomic field. Until now, deep learning has been widely used in the functional analysis of DNA sequences, including DeepSEA, DanQ, DeepATT and TBiNet. However, these methods have the problems of high computational complexity and not fully considering the distant interactions among chromatin features, thus affecting the prediction accuracy. In this work, we propose a hybrid deep neural network model, called DeepFormer, based on convolutional neural network (CNN) and flow-attention mechanism for DNA sequence function prediction. In DeepFormer, the CNN is used to capture the local features of DNA sequences as well as important motifs. Based on the conservation law of flow network, the flow-attention mechanism can capture more distal interactions among sequence features with linear time complexity. We compare DeepFormer with the above four kinds of classical methods using the commonly used dataset of 919 chromatin features of nearly 4.9 million noncoding DNA sequences. Experimental results show that DeepFormer significantly outperforms four kinds of methods, with an average recall rate at least 7.058% higher than other methods. Furthermore, we confirmed the effectiveness of DeepFormer in capturing functional variation using Alzheimer's disease, pathogenic mutations in alpha-thalassemia and modification in CCCTC-binding factor (CTCF) activity. We further predicted the maize chromatin accessibility of five tissues and validated the generalization of DeepFormer. The average recall rate of DeepFormer exceeds the classical methods by at least 1.54%, demonstrating strong robustness.

YTH Domain Proteins Play an Essential Role in Rice Growth and Stress Response

As the most prevalent epi-transcriptional modification, m6A modifications play essential roles in regulating RNA fate. The molecular functions of YTH521-B homology (YTH) domain proteins, the most known READER proteins of m6A modifications, have been well-studied in animals. Although plants contain more YTH domain proteins than other eukaryotes, little is known about their biological importance. In dicot species Arabidopsis thaliana, the YTHDFA clade members ECT2/3/4 and CPSF30-L are well-studied and important for cell proliferation, plant organogenesis, and nitrate transport. More emphasis is needed on the biological functions of plant YTH proteins, especially monocot YTHs. Here we presented a detailed phylogenetic relationship of eukaryotic YTH proteins and clustered plant YTHDFC clade into three subclades. To determine the importance of monocot YTH proteins, YTH knockout mutants and RNAi-induced knockdown plants were constructed and used for phenotyping, transcriptomic analysis, and stress treatments. Knocking out or knocking down OsYTHs led to the downregulation of multicellular organismal regulation genes and resulted in growth defects. In addition, loss-of-function ythdfa mutants led to better salinity tolerance whereas ythdfc mutants were more sensitive to abiotic stress. Overall, our study establishes the functional relevance of rice YTH genes in plant growth regulation and stress response.

NF-κB signaling controls H3K9me3 levels at intronic LINE-1 and hematopoietic stem cell genes in cis

Ionizing radiations (IR) alter hematopoietic stem cell (HSC) function on the long term, but the mechanisms underlying these effects are still poorly understood. We recently showed that IR induces the derepression of L1Md, the mouse young subfamilies of LINE-1/L1 retroelements. L1 contributes to gene regulatory networks. However, how L1Md are derepressed and impact HSC gene expression are not known. Here, we show that IR triggers genome-wide H3K9me3 decrease that occurs mainly at L1Md. Loss of H3K9me3 at intronic L1Md harboring NF-κB binding sites motifs but not at promoters is associated with the repression of HSC-specific genes. This is correlated with reduced NFKB1 repressor expression. TNF-α treatment rescued all these effects and prevented IR-induced HSC loss of function in vivo. This TNF-α/NF-κB/H3K9me3/L1Md axis might be important to maintain HSCs while allowing expression of immune genes during myeloid regeneration or damage-induced bone marrow ablation.

Venus: Elucidating the Impact of Amino Acid Variants on Protein Function Beyond Structure Destabilisation

Exploring the functional effect of a non-synonymous coding variant at the protein level requires multiple pieces of information to be interpreted appropriately. This is particularly important when embarking on the study of a potentially pathogenic variant linked to a rare or monogenic disease. Whereas accurate protein stability predictions alone are generally informative, other effects, such as disruption of post-translational modifications or weakened ligand binding, may also contribute to the disease phenotype. Furthermore, consideration of nearby variants that are found in the healthy population may strengthen or refute a given mechanistic hypothesis. Whilst there are several bioinformatics tools available that score a genetic variant in terms of deleteriousness, there is no single tool that assembles multiple effects of a variant on the encoded protein, beyond structural stability, and presents them on the structure for inspection. Venus is a web application which, given a protein substitution, rapidly estimates the predicted effect on protein stability of the variant, flags if the variant affects a post-translational modification site, a predicted linear motif or known annotation, and determines the effect on protein stability of variants which affect nearby residues and have been identified in healthy populations. Venus is built upon Michelanglo and the results can be exported to it, allowing them to be annotated and shared with other researchers. Venus is freely accessible at https://venus.cmd.ox.ac.uk and its source code is openly available at https://github.com/CMD-Oxford/Michelanglo-and-Venus.

Machine learning: its challenges and opportunities in plant system biology

Sequencing technologies are evolving at a rapid pace, enabling the generation of massive amounts of data in multiple dimensions (e.g., genomics, epigenomics, transcriptomic, metabolomics, proteomics, and single-cell omics) in plants. To provide comprehensive insights into the complexity of plant biological systems, it is important to integrate different omics datasets. Although recent advances in computational analytical pipelines have enabled efficient and high-quality exploration and exploitation of single omics data, the integration of multidimensional, heterogenous, and large datasets (i.e., multi-omics) remains a challenge. In this regard, machine learning (ML) offers promising approaches to integrate large datasets and to recognize fine-grained patterns and relationships. Nevertheless, they require rigorous optimizations to process multi-omics-derived datasets. In this review, we discuss the main concepts of machine learning as well as the key challenges and solutions related to the big data derived from plant system biology. We also provide in-depth insight into the principles of data integration using ML, as well as challenges and opportunities in different contexts including multi-omics, single-cell omics, protein function, and protein-protein interaction. KEY POINTS: • The key challenges and solutions related to the big data derived from plant system biology have been highlighted. • Different methods of data integration have been discussed. • Challenges and opportunities of the application of machine learning in plant system biology have been highlighted and discussed.

Annotating functional effects of non-coding variants in neuropsychiatric cell types by deep transfer learning

Genomewide association studies (GWAS) have identified a large number of loci associated with neuropsychiatric traits, however, understanding the molecular mechanisms underlying these loci remains difficult. To help prioritize causal variants and interpret their functions, computational methods have been developed to predict regulatory effects of non-coding variants. An emerging approach to variant annotation is deep learning models that predict regulatory functions from DNA sequences alone. While such models have been trained on large publicly available dataset such as ENCODE, neuropsychiatric trait-related cell types are under-represented in these datasets, thus there is an urgent need of better tools and resources to annotate variant functions in such cellular contexts. To fill this gap, we collected a large collection of neurodevelopment-related cell/tissue types, and trained deep Convolutional Neural Networks (ResNet) using such data. Furthermore, our model, called MetaChrom, borrows information from public epigenomic consortium to improve the accuracy via transfer learning. We show that MetaChrom is substantially better in predicting experimentally determined chromatin accessibility variants than popular variant annotation tools such as CADD and delta-SVM. By combining GWAS data with MetaChrom predictions, we prioritized 31 SNPs for Schizophrenia, suggesting potential risk genes and the biological contexts where they act. In summary, MetaChrom provides functional annotations of any DNA variants in the neuro-development context and the general method of MetaChrom can also be extended to other disease-related cell or tissue types.

Machine learning meets omics: applications and perspectives

The innovation of biotechnologies has allowed the accumulation of omics data at an alarming rate, thus introducing the era of 'big data'. Extracting inherent valuable knowledge from various omics data remains a daunting problem in bioinformatics. Better solutions often need some kind of more innovative methods for efficient handlings and effective results. Recent advancements in integrated analysis and computational modeling of multi-omics data helped address such needs in an increasingly harmonious manner. The development and application of machine learning have largely advanced our insights into biology and biomedicine and greatly promoted the development of therapeutic strategies, especially for precision medicine. Here, we propose a comprehensive survey and discussion on what happened, is happening and will happen when machine learning meets omics. Specifically, we describe how artificial intelligence can be applied to omics studies and review recent advancements at the interface between machine learning and the ever-widest range of omics including genomics, transcriptomics, proteomics, metabolomics, radiomics, as well as those at the single-cell resolution. We also discuss and provide a synthesis of ideas, new insights, current challenges and perspectives of machine learning in omics.

Machine learning applications for therapeutic tasks with genomics data

Thanks to the increasing availability of genomics and other biomedical data, many machine learning algorithms have been proposed for a wide range of therapeutic discovery and development tasks. In this survey, we review the literature on machine learning applications for genomics through the lens of therapeutic development. We investigate the interplay among genomics, compounds, proteins, electronic health records, cellular images, and clinical texts. We identify 22 machine learning in genomics applications that span the whole therapeutics pipeline, from discovering novel targets, personalizing medicine, developing gene-editing tools, all the way to facilitating clinical trials and post-market studies. We also pinpoint seven key challenges in this field with potentials for expansion and impact. This survey examines recent research at the intersection of machine learning, genomics, and therapeutic development.

Ten Years of EWAS

Epigenome-wide association study (EWAS) has been applied to analyze DNA methylation variation in complex diseases for a decade, and epigenome as a research target has gradually become a hot topic of current studies. The DNA methylation microarrays, next-generation, and third-generation sequencing technologies have prepared a high-quality platform for EWAS. Here, the progress of EWAS research is reviewed, its contributions to clinical applications, and mainly describe the achievements of four typical diseases. Finally, the challenges encountered by EWAS and make bold predictions for its future development are presented.

Accurate prediction of cis-regulatory modules reveals a prevalent regulatory genome of humans

cis-regulatory modules(CRMs) formed by clusters of transcription factor (TF) binding sites (TFBSs) are as important as coding sequences in specifying phenotypes of humans. It is essential to categorize all CRMs and constituent TFBSs in the genome. In contrast to most existing methods that predict CRMs in specific cell types using epigenetic marks, we predict a largely cell type agonistic but more comprehensive map of CRMs and constituent TFBSs in the gnome by integrating all available TF ChIP-seq datasets. Our method is able to partition 77.47% of genome regions covered by available 6092 datasets into a CRM candidate (CRMC) set (56.84%) and a non-CRMC set (43.16%). Intriguingly, the predicted CRMCs are under strong evolutionary constraints, while the non-CRMCs are largely selectively neutral, strongly suggesting that the CRMCs are likely cis-regulatory, while the non-CRMCs are not. Our predicted CRMs are under stronger evolutionary constraints than three state-of-the-art predictions (GeneHancer, EnhancerAtlas and ENCODE phase 3) and substantially outperform them for recalling VISTA enhancers and non-coding ClinVar variants. We estimated that the human genome might encode about 1.47M CRMs and 68M TFBSs, comprising about 55% and 22% of the genome, respectively; for both of which, we predicted 80%. Therefore, the cis-regulatory genome appears to be more prevalent than originally thought.

Learning the Regulatory Code of Gene Expression

Data-driven machine learning is the method of choice for predicting molecular phenotypes from nucleotide sequence, modeling gene expression events including protein-DNA binding, chromatin states as well as mRNA and protein levels. Deep neural networks automatically learn informative sequence representations and interpreting them enables us to improve our understanding of the regulatory code governing gene expression. Here, we review the latest developments that apply shallow or deep learning to quantify molecular phenotypes and decode the cis-regulatory grammar from prokaryotic and eukaryotic sequencing data. Our approach is to build from the ground up, first focusing on the initiating protein-DNA interactions, then specific coding and non-coding regions, and finally on advances that combine multiple parts of the gene and mRNA regulatory structures, achieving unprecedented performance. We thus provide a quantitative view of gene expression regulation from nucleotide sequence, concluding with an information-centric overview of the central dogma of molecular biology.

Incorporating K-mers Highly Correlated to Epigenetic Modifications for Bayesian Inference of Gene Interactions

Objective::

A gene interaction network, along with its related biological features, has an important role in computational biology. Bayesian network, as an efficient model, based on probabilistic concepts is able to exploit known and novel biological casual relationships between genes. The success of Bayesian networks in predicting the relationships greatly depends on selecting priors.

Methods::

K-mers have been applied as the prominent features to uncover the similarity between genes in a specific pathway, suggesting that this feature can be applied to study genes dependencies. In this study, we propose k-mers (4,5 and 6-mers) highly correlated with epigenetic modifications, including 17 modifications, as a new prior for Bayesian inference in the gene interaction network.

Result::

Employing this model on a network of 23 human genes and on a network based on 27 genes related to yeast resulted in F-measure improvements in different biological networks.

Conclusion::

The improvements in the best case are 12%, 36%, and 10% in the pathway, coexpression, and physical interaction, respectively.

CpG content-dependent associations between transcription factors and histone modifications

Understanding the factors that underlie the epigenetic regulation of genes is crucial to understand the gene regulatory machinery as a whole. Several experimental and computational studies examined the relationship between different factors involved. Here we investigate the relationship between transcription factors (TFs) and histone modifications (HMs), based on ChIP-seq data in cell lines. As it was shown that gene regulation by TFs differs depending on the CpG class of a promoter, we study the impact of the CpG content in promoters on the associations between TFs and HMs. We suggest an approach based on sparse linear regression models to infer associations between TFs and HMs with respect to CpG content. A study of the partial correlation of HMs for the two classes of high and low CpG content reveals possible CpG dependence and potential candidates for confounding factors in our models. We show that the models are accurate, inferred associations reflect known biological relationships, and we give new insight into associations with respect to CpG content. Moreover, analysis of a ChIP-seq dataset in HepG2 cells of the HM H3K122ac, an HM about little is known, reveals novel TF associations and supports a previously established link to active transcription.

The impact of flanking sequence features on DNA CpG methylation

Epigenetics and DNA methylation play a pivotal role in many processes of the cell and we often observe that an aberrant methylation pattern characterizes pathologies. In this work we investigate the role that the flanking sequences of CGs play in the methylation process in human. We built four different CG datasets: methylated, unmethylated, and two randomly extracted ones. We evaluated features associated to the flanking sequences of those CG sets, for different size around the CG, through five measures accounting for different aspects of sequence composition complexity and structure. The analysis performed through those measures revealed evident different behaviors between methylated and unmethylated probe sets. Major differences were observed for GC content and CG dinucleotide frequency in a window size of 300-400 bp and for CG self-attraction in 3K bp. It is remarkable as the effect of methylated CG lasts much more than expected far from the CG.

Single-molecule multiplexed profiling of protein-DNA complexes using magnetic tweezers

Epigenetics, such as the dynamic interplay between DNA methylation and demethylation, play diverse roles in critical cellular events. Enzymatic activity at CpG sites, where cytosines are methylated or demethylated, is known to be influenced by the density of CpGs, methylation states, and the flanking sequences of a CpG site. However, how the relevant enzymes are recruited to and recognize their target DNA is less clear. Moreover, although DNA-binding epigenetic enzymes are ideal targets for therapeutic intervention, these targets have been rarely exploited. Single-molecule techniques offer excellent capabilities to probe site-specific protein-DNA interactions and unravel the dynamics. Here, we develop a single-molecule approach that allows multiplexed profiling of protein-DNA complexes using magnetic tweezers. When a DNA hairpin with multiple binding sites is unzipping, strand separation pauses at the positions bound by a protein. We can thus measure site-specific binding probabilities and dissociation time directly. Taking the TET1 CXXC domain as an example, we show that TET1 CXXC binds multiple CpG motifs with various flanking nucleotides or different methylation patterns in an AT-rich DNA. We are able to establish for the first time, at nanometer resolution, that TET1 CXXC prefers G/C flanked CpG motif over C/G, A/T, or T/A flanked ones. CpG methylation strengthens TET1 CXXC recruitment but has little effect on dissociation time. Finally, we demonstrate that TET1 CXXC can distinguish five CpG clusters in a CpG island with crowded binding motifs. We anticipate that the feasibility of single-molecule multiplexed profiling assays will contribute to the understanding of protein-DNA interactions.

Deciphering the genetic code of DNA methylation

DNA methylation plays crucial roles in many biological processes and abnormal DNA methylation patterns are often observed in diseases. Recent studies have shed light on cis-acting DNA elements that regulate locus-specific DNA methylation, which involves transcription factors, histone modification and DNA secondary structures. In addition, several recent studies have surveyed DNA motifs that regulate DNA methylation and suggest potential applications in diagnosis and prognosis. Here, we discuss the current biological foundation for the cis-acting genetic code that regulates DNA methylation. We review the computational models that predict DNA methylation with genetic features and discuss the biological insights revealed from these models. We also provide an in-depth discussion on how to leverage such knowledge in clinical applications, particularly in the context of liquid biopsy for early cancer diagnosis and treatment.

Parsing the Functional Impact of Noncoding Genetic Variants in the Brain Epigenome

The heritability of common psychiatric disorders has motivated global efforts to identify risk-associated genetic variants and elucidate molecular pathways connecting DNA sequence to disease-associated brain dysfunction. The overrepresentation of risk variants among gene regulatory loci instead of protein-coding loci, however, poses a unique challenge in discerning which among the many thousands of variants identified contribute functionally to disease etiology. Defined broadly, psychiatric epigenomics seeks to understand the effects of disease-associated genetic variation on functional readouts of chromatin in an effort to prioritize variants in terms of their impact on gene expression in the brain. Here, we provide an overview of epigenomic mapping in the human brain and highlight findings of particular relevance to psychiatric genetics. Computational methods, including convolutional neuronal networks, and other machine learning approaches hold great promise for elucidating the functional impact of both common and rare genetic variants, thereby refining the epigenomic architecture of psychiatric disorders and enabling integrative analyses of regulatory noncoding variants in the context of large population-level genome and phenome databases.

Learning and interpreting the gene regulatory grammar in a deep learning framework

Deep neural networks (DNNs) have achieved state-of-the-art performance in identifying gene regulatory sequences, but they have provided limited insight into the biology of regulatory elements due to the difficulty of interpreting the complex features they learn. Several models of how combinatorial binding of transcription factors, i.e. the regulatory grammar, drives enhancer activity have been proposed, ranging from the flexible TF billboard model to the stringent enhanceosome model. However, there is limited knowledge of the prevalence of these (or other) sequence architectures across enhancers. Here we perform several hypothesis-driven analyses to explore the ability of DNNs to learn the regulatory grammar of enhancers. We created synthetic datasets based on existing hypotheses about combinatorial transcription factor binding site (TFBS) patterns, including homotypic clusters, heterotypic clusters, and enhanceosomes, from real TF binding motifs from diverse TF families. We then trained deep residual neural networks (ResNets) to model the sequences under a range of scenarios that reflect real-world multi-label regulatory sequence prediction tasks. We developed a gradient-based unsupervised clustering method to extract the patterns learned by the ResNet models. We demonstrated that simulated regulatory grammars are best learned in the penultimate layer of the ResNets, and the proposed method can accurately retrieve the regulatory grammar even when there is heterogeneity in the enhancer categories and a large fraction of TFBS outside of the regulatory grammar. However, we also identify common scenarios where ResNets fail to learn simulated regulatory grammars. Finally, we applied the proposed method to mouse developmental enhancers and were able to identify the components of a known heterotypic TF cluster. Our results provide a framework for interpreting the regulatory rules learned by ResNets, and they demonstrate that the ability and efficiency of ResNets in learning the regulatory grammar depends on the nature of the prediction task.

Motto: Representing Motifs in Consensus Sequences with Minimum Information Loss

Sequence analysis frequently requires intuitive understanding and convenient representation of motifs. Typically, motifs are represented as position weight matrices (PWMs) and visualized using sequence logos. However, in many scenarios, in order to interpret the motif information or search for motif matches, it is compact and sufficient to represent motifs by wildcard-style consensus sequences (such as [GC][AT]GATAAG[GAC]). Based on mutual information theory and Jensen-Shannon divergence, we propose a mathematical framework to minimize the information loss in converting PWMs to consensus sequences. We name this representation as sequence Motto and have implemented an efficient algorithm with flexible options for converting motif PWMs into Motto from nucleotides, amino acids, and customized characters. We show that this representation provides a simple and efficient way to identify the binding sites of 1156 common transcription factors (TFs) in the human genome. The effectiveness of the method was benchmarked by comparing sequence matches found by Motto with PWM scanning results found by FIMO. On average, our method achieves a 0.81 area under the precision-recall curve, significantly (P-value < 0.01) outperforming all existing methods, including maximal positional weight, Cavener's method, and minimal mean square error. We believe this representation provides a distilled summary of a motif, as well as the statistical justification.

Evaluating the informativeness of deep learning annotations for human complex diseases

Deep learning models have shown great promise in predicting regulatory effects from DNA sequence, but their informativeness for human complex diseases is not fully understood. Here, we evaluate genome-wide SNP annotations from two previous deep learning models, DeepSEA and Basenji, by applying stratified LD score regression to 41 diseases and traits (average N = 320K), conditioning on a broad set of coding, conserved and regulatory annotations. We aggregated annotations across all (respectively blood or brain) tissues/cell-types in meta-analyses across all (respectively 11 blood or 8 brain) traits. The annotations were highly enriched for disease heritability, but produced only limited conditionally significant results: non-tissue-specific and brain-specific Basenji-H3K4me3 for all traits and brain traits respectively. We conclude that deep learning models have yet to achieve their full potential to provide considerable unique information for complex disease, and that their conditional informativeness for disease cannot be inferred from their accuracy in predicting regulatory annotations.

Deep learning models have shown great promise in predicting regulatory effects from DNA sequence. Here the authors evaluate sequence-based epigenomic deep learning models and conclude that these models are not yet ready to inform our knowledge of human disease.

Predicting DNA Methylation States with Hybrid Information Based Deep-Learning Model

DNA methylation plays an important role in the regulation of some biological processes. Up to now, with the development of machine learning models, there are several sequence-based deep learning models designed to predict DNA methylation states, which gain better performance than traditional methods like random forest and SVM. However, convolutional network based deep learning models that use one-hot encoding DNA sequence as input may discover limited information and cause unsatisfactory prediction performance, so more data and model structures of diverse angles should be considered. In this work, we proposed a hybrid sequence-based deep learning model with both MeDIP-seq data and Histone information to predict DNA methylated CpG states (MHCpG). We combined both MeDIP-seq data and histone modification data with sequence information and implemented convolutional network to discover sequence patterns. In addition, we used statistical data gained from previous three input data and adopted a 3-layer feedforward neuron network to extract more high-level features. We compared our method with traditional predicting methods using random forest and other previous methods like CpGenie and DeepCpG, the result showed that MHCpG exceeded the other approaches and gained more satisfactory performance.

DeepATT: a hybrid category attention neural network for identifying functional effects of DNA sequences

Quantifying DNA properties is a challenging task in the broad field of human genomics. Since the vast majority of non-coding DNA is still poorly understood in terms of function, this task is particularly important to have enormous benefit for biology research. Various DNA sequences should have a great variety of representations, and specific functions may focus on corresponding features in the front part of learning model. Currently, however, for multi-class prediction of non-coding DNA regulatory functions, most powerful predictive models do not have appropriate feature extraction and selection approaches for specific functional effects, so that it is difficult to gain a better insight into their internal correlations. Hence, we design a category attention layer and category dense layer in order to select efficient features and distinguish different DNA functions. In this study, we propose a hybrid deep neural network method, called DeepATT, for identifying $919$ regulatory functions on nearly $5$ million DNA sequences. Our model has four built-in neural network constructions: convolution layer captures regulatory motifs, recurrent layer captures a regulatory grammar, category attention layer selects corresponding valid features for different functions and category dense layer classifies predictive labels with selected features of regulatory functions. Importantly, we compare our novel method, DeepATT, with existing outstanding prediction tools, DeepSEA and DanQ. DeepATT performs significantly better than other existing tools for identifying DNA functions, at least increasing $1.6\%$ area under precision recall. Furthermore, we can mine the important correlation among different DNA functions according to the category attention module. Moreover, our novel model can greatly reduce the number of parameters by the mechanism of attention and locally connected, on the basis of ensuring accuracy.

Prediction of RNA Methylation Status From Gene Expression Data Using Classification and Regression Methods

RNA N⁶-methyladenosine (m⁶A) has emerged as an important epigenetic modification for its role in regulating the stability, structure, processing, and translation of RNA. Instability of m⁶A homeostasis may result in flaws in stem cell regulation, decrease in fertility, and risk of cancer. To this day, experimental detection and quantification of RNA m⁶A modification are still time-consuming and labor-intensive. There is only a limited number of epitranscriptome samples in existing databases, and a matched RNA methylation profile is not often available for a biological problem of interests. As gene expression data are usually readily available for most biological problems, it could be appealing if we can estimate the RNA methylation status from gene expression data using in silico methods. In this study, we explored the possibility of computational prediction of RNA methylation status from gene expression data using classification and regression methods based on mouse RNA methylation data collected from 73 experimental conditions. Elastic Net-regularized Logistic Regression (ENLR), Support Vector Machine (SVM), and Random Forests (RF) were constructed for classification. Both SVM and RF achieved the best performance with the mean area under the curve (AUC) = 0.84 across samples; SVM had a narrower AUC spread. Gene Site Enrichment Analysis was conducted on those sites selected by ENLR as predictors to access the biological significance of the model. Three functional annotation terms were found statistically significant: phosphoprotein, SRC Homology 3 (SH3) domain, and endoplasmic reticulum. All 3 terms were found to be closely related to m⁶A pathway. For regression analysis, Elastic Net was implemented, which yielded a mean Pearson correlation coefficient = 0.68 and a mean Spearman correlation coefficient = 0.64. Our exploratory study suggested that gene expression data could be used to construct predictors for m⁶A methylation status with adequate accuracy. Our work showed for the first time that RNA methylation status may be predicted from the matched gene expression data. This finding may facilitate RNA modification research in various biological contexts when a matched RNA methylation profile is not available, especially in the very early stage of the study.

OncoBase: a platform for decoding regulatory somatic mutations in human cancers

Whole-exome and whole-genome sequencing have revealed millions of somatic mutations associated with different human cancers, and the vast majority of them are located outside of coding sequences, making it challenging to directly interpret their functional effects. With the rapid advances in high-throughput sequencing technologies, genome-scale long-range chromatin interactions were detected, and distal target genes of regulatory elements were determined using three-dimensional (3D) chromatin looping. Herein, we present OncoBase (http://www.oncobase.biols.ac.cn/), an integrated database for annotating 81 385 242 somatic mutations in 68 cancer types from more than 120 cancer projects by exploring their roles in distal interactions between target genes and regulatory elements. OncoBase integrates local chromatin signatures, 3D chromatin interactions in different cell types and reconstruction of enhancer-target networks using state-of-the-art algorithms. It employs informative visualization tools to display the integrated local and 3D chromatin signatures and effects of somatic mutations on regulatory elements. Enhancer-promoter interactions estimated from chromatin interactions are integrated into a network diffusion system that quantitatively prioritizes somatic mutations and target genes from a large pool. Thus, OncoBase is a useful resource for the functional annotation of regulatory noncoding regions and systematically benchmarking the regulatory effects of embedded noncoding somatic mutations in human carcinogenesis.

ICGEC: a comparative method for measuring epigenetic conservation of genes via the integrated signal from multiple histone modifications between cell types

Background

Histone post-translational modifications play crucial roles in epigenetic regulation of gene expression and are known to be associated with the phenotypic differences of different cell types. Therefore, it is of fundamental importance to dissect the genes and pathways involved in such a phenotypic variation at the level of epigenetics. However, the existing comparative approaches are largely based on the differences, especially the absolute difference in the levels of individual histone modifications of genes under contrasting conditions. Thus, a method for measuring the overall change in the epigenetic circumstance of each gene underpinned by multiple types of histone modifications between cell types is lacking.

Results

To address this challenge, we developed ICGEC, a new method for estimating the degree of epigenetic conservation of genes between two cell lines. Different from existing comparative methods, ICGEC provides a reliable score for measuring the relative change in the epigenetic context of corresponding gene between two conditions and simultaneously produces a score for each histone mark. The application of ICGEC to the human embryonic stem cell line H1 and four H1-derived cell lines with available epigenomic data for the same 16 types of histone modifications indicated high robustness and reliability of ICGEC. Furthermore, the analysis of the epigenetically dynamic and conserved genes which were defined based on the ICGEC output results demonstrated that ICGEC can deepen our understanding of the biological processes of cell differentiation to overcome the limitations of traditional expression analysis. Specifically, the ICGEC-derived differentiation-direction-specific genes were shown to have putative functions that are well-matched with cell identity. Additionally, H3K79me1 and H3K27ac were found to be the main histone marks accounting for whether an epigenetically dynamic gene was differentially expressed between two cell lines.

Conclusions

The use of ICGEC creates a convenient and robust way to measure the overall epigenetic conservation of individual genes and marks between two conditions. Thus, it provides a basis for exploring the epigenotype-phenotype relationship. ICGEC can be deemed a state-of-the-art method tailored for comparative epigenomic analysis of changes in cell dynamics.

SETDB1 promotes the progression of colorectal cancer via epigenetically silencing p21 expression

SETDB1, a histone H3K9 methyltransferase, has been reported to be upregulated in a variety of tumors and promotes cancer development. However, the exact pathogenesis of SETDB1 in human colorectal cancer (CRC) is hitherto unknown. Here, we showed that SETDB1 expression was highly amplified in CRC. Functionally, SETDB1 downregulation in SW480 and HCT116 cells reduced cell proliferation, migration, invasion, and increased CRC cells apoptosis. In contrast, SETDB1 overexpression promoted CRC cells proliferation, migration, and invasion. High expression of SETDB1 was associated with a more aggressive phenotype in vitro. Flow cytometry showed that cell cycle was arrested in G1 phase after SETDB1 silencing. Furthermore, depletion of SETDB1 in vivo suppressed CRC cells proliferation. Mechanistically, p21 was identified as the target of SETDB1. After transfected with siSETDB1, expression of p21 was distinctly increased. In contrast, expression of p21 was significantly decreased after overexpression SETDB1. We also showed that SETDB1 could be involved in the regulation of epithelial–mesenchymal transition (EMT) in HCT116 cells. Moreover, we confirmed that SETDB1 could regulate the activity of p21 promoter by dual-luciferase repoter assay, and proved that SETDB1 could bind to the promoter of p21 and regulate its H3K9me3 enrichment level by ChIP-PCR experiment. Finally, we verified that silencing of SETDB1 inhibited CRC tumorigenesis in vivo. In conclusion, our results indicate that SETDB1 is a major driver of CRC development and might provide a new therapeutic target for the clinical treatment of CRC.

ncHMR detector: a computational framework to systematically reveal non-classical functions of histone modification regulators

Recently, several non-classical functions of histone modification regulators (HMRs), independent of their known histone modification substrates and products, have been reported to be essential for specific cellular processes. However, there is no framework designed for identifying such functions systematically. Here, we develop ncHMR detector, the first computational framework to predict non-classical functions and cofactors of a given HMR, based on ChIP-seq data mining. We apply ncHMR detector in ChIP-seq data-rich cell types and predict non-classical functions of HMRs. Finally, we experimentally reveal that the predicted non-classical function of CBX7 is biologically significant for the maintenance of pluripotency.

A comprehensive review of computational prediction of genome-wide features

There are significant correlations among different types of genetic, genomic and epigenomic features within the genome. These correlations make the in silico feature prediction possible through statistical or machine learning models. With the accumulation of a vast amount of high-throughput data, feature prediction has gained significant interest lately, and a plethora of papers have been published in the past few years. Here we provide a comprehensive review on these published works, categorized by the prediction targets, including protein binding site, enhancer, DNA methylation, chromatin structure and gene expression. We also provide discussions on some important points and possible future directions.

MultiMotifMaker: A Multi-Thread Tool for Identifying DNA Methylation Motifs from Pacbio Reads

The methylation of DNA is an important mechanism to control biological processes. Recently, the Pacbio SMRT technology provides a new way to identify base methylation in the genome. MotifMaker is a tool developed by Pacbio for discovering DNA methylation motifs from methylated DNA sequences. However, MotifMaker is single-threaded and computational expensive for identifying methylation motifs from large genomes. Here, we present an efficient motif finding algorithm (MultiMotifMaker) by implementing multi threads of the MotifMaker. The MultiMotifMaker speeds up the motif search about 8-9 times on a 32 core computer comparing to MotifMaker. MultiMotifMaker makes it possible to identify methylation motifs from Pacbio reads for large genomes.

Machine Learning for Integrating Data in Biology and Medicine: Principles, Practice, and Opportunities

New technologies have enabled the investigation of biology and human health at an unprecedented scale and in multiple dimensions. These dimensions include myriad properties describing genome, epigenome, transcriptome, microbiome, phenotype, and lifestyle. No single data type, however, can capture the complexity of all the factors relevant to understanding a phenomenon such as a disease. Integrative methods that combine data from multiple technologies have thus emerged as critical statistical and computational approaches. The key challenge in developing such approaches is the identification of effective models to provide a comprehensive and relevant systems view. An ideal method can answer a biological or medical question, identifying important features and predicting outcomes, by harnessing heterogeneous data across several dimensions of biological variation. In this Review, we describe the principles of data integration and discuss current methods and available implementations. We provide examples of successful data integration in biology and medicine. Finally, we discuss current challenges in biomedical integrative methods and our perspective on the future development of the field.

Finding de novo methylated DNA motifs

MOTIVATION

Increasing evidence has shown that nucleotide modifications such as methylation and hydroxymethylation on cytosine would greatly impact the binding of transcription factors (TFs). However, there is a lack of motif finding algorithms with the function to search for motifs with modified bases. In this study, we expand on our previous motif finding pipeline Epigram to provide systematic de novo motif discovery and performance evaluation on methylated DNA motifs.

RESULTS

mEpigram outperforms both MEME and DREME on finding modified motifs in simulated data that mimics various motif enrichment scenarios. Furthermore we were able to identify methylated motifs in Arabidopsis DNA affinity purification sequencing (DAP-seq) data that were previously demonstrated to contain such motifs. When applied to TF ChIP-seq and DNA methylome data in H1 and GM12878, our method successfully identified novel methylated motifs that can be recognized by the TFs or their co-factors. We also observed spacing constraint between the canonical motif of the TF and the newly discovered methylated motifs, which suggests operative recognition of these cis-elements by collaborative proteins.

AVAILABILITY AND IMPLEMENTATION

The mEpigram program is available at http://wanglab.ucsd.edu/star/mEpigram.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Deciphering epigenomic code for cell differentiation using deep learning

Background

Although DNA sequence plays a crucial role in establishing the unique epigenome of a cell type, little is known about the sequence determinants that lead to the unique epigenomes of different cell types produced during cell differentiation. To fill this gap, we employed two types of deep convolutional neural networks (CNNs) constructed for each of differentially related cell types and for each of histone marks measured in the cells, to learn the sequence determinants of various histone modification patterns in each cell type.

Results

We applied our models to four differentially related human CD₄⁺ T cell types and six histone marks measured in each cell type. The cell models can accurately predict the histone marks in each cell type, while the mark models can also accurately predict the cell types based on a single mark. Sequence motifs learned by both the cell or mark models are highly similar to known binding motifs of transcription factors known to play important roles in CD₄⁺ T cell differentiation. Both the unique histone mark patterns in each cell type and the different patterns of the same histone mark in different cell types are determined by a set of motifs with unique combinations. Interestingly, the level of sharing motifs learned in the different cell models reflects the lineage relationships of the cells, while the level of sharing motifs learned in the different histone mark models reflects their functional relationships. These models can also enable the prediction of the importance of learned motifs and their interactions in determining specific histone mark patterns in the cell types.

Conclusion

Sequence determinants of various histone modification patterns in different cell types can be revealed by comparative analysis of motifs learned in the CNN models for multiple cell types and histone marks. The learned motifs are interpretable and may provide insights into the underlying molecular mechanisms of establishing the unique epigenomes in different cell types. Thus, our results support the hypothesis that DNA sequences ultimately determine the unique epigenomes of different cell types through their interactions with transcriptional factors, epigenome remodeling system and extracellular cues during cell differentiation.

Electronic supplementary material

The online version of this article (10.1186/s12864-019-6072-8) contains supplementary material, which is available to authorized users.

Identification of DNA motifs that regulate DNA methylation

DNA methylation is an important epigenetic mark but how its locus-specificity is decided in relation to DNA sequence is not fully understood. Here, we have analyzed 34 diverse whole-genome bisulfite sequencing datasets in human and identified 313 motifs, including 92 and 221 associated with methylation (methylation motifs, MMs) and unmethylation (unmethylation motifs, UMs), respectively. The functionality of these motifs is supported by multiple lines of evidence. First, the methylation levels at the MM and UM motifs are respectively higher and lower than the genomic background. Second, these motifs are enriched at the binding sites of methylation modifying enzymes including DNMT3A and TET1, indicating their possible roles of recruiting these enzymes. Third, these motifs significantly overlap with "somatic QTLs" (quantitative trait loci) of methylation and expression. Fourth, disruption of these motifs by mutation is associated with significantly altered methylation level of the CpGs in the neighbor regions. Furthermore, these motifs together with somatic mutations are predictive of cancer subtypes and patient survival. We revealed some of these motifs were also associated with histone modifications, suggesting a possible interplay between the two types of epigenetic modifications. We also found some motifs form feed forward loops to contribute to DNA methylation dynamics.

Chromatin Signature and Transcription Factor Binding Provide a Predictive Basis for Understanding Plant Gene Expression

Chromatin accessibility and post-transcriptional histone modifications play important roles in gene expression regulation. However, little is known about the joint effect of multiple chromatin modifications on the gene expression level in plants, despite that the regulatory roles of individual histone marks such as H3K4me3 in gene expression have been well-documented. By using machine-learning methods, we systematically performed gene expression level prediction based on multiple chromatin modifications data in Arabidopsis and rice. We found that as few as four histone modifications were sufficient to yield good prediction performance, and H3K4me3 and H3K36me3 being the top two predictors with known functions related to transcriptional initiation and elongation, respectively. We demonstrated that the predictive powers differed between protein-coding and non-coding genes as well as between CpG-enriched and CpG-depleted genes. We also showed that the predictive model trained in one tissue or species could be applied to another tissue or species, suggesting shared underlying mechanisms. More interestingly, the gene expression levels of conserved orthologs are easier to predict than the species-specific genes. In addition, chromatin state of distal enhancers was moderately correlated to gene expression but was dispensable if given the chromatin features of the proximal regions of genes. We further extended the analysis to transcription factor (TF) binding data. Strikingly, the combinatorial effects of only a few TFs were roughly fit to gene expression levels in Arabidopsis. Overall, by using quantitative modeling, we provide a comprehensive and unbiased perspective on the epigenetic and TF-mediated regulation of gene expression in plants.

Genetic Contexts Characterize Dynamic Histone Modification Patterns Among Cell Types

Histone modifications play critical roles in mammalian development, regulating chromatin structure and gene expression. Dynamic histone modifications among cell types have been shown to associate with changes in mammalian development. However, how to quantitatively measure the histone modification alterations and how histone modifications vary across cell types under different genetic contexts remain largely unexplored and whether these changes are related to the primary DNA sequence remains limited. Here, we employed an entropy-based method to measure histone modification alterations in six definite genomic regions across five cell types and identified lineage-specific histone modification genes. We observed that histone modification alterations prefer to enrich in 5'-UTR exons, and also in 3'-UTR exons and its downstream. Then we built a model to predict the histone modification patterns from the primary DNA sequence. We found that the frequencies of k-mer sequence compositions are predictive of histone modification patterns, suggesting that the primary DNA sequence correlated with the histone modification alterations among cell types. Additionally, the lineage-specific histone modification genes display a higher conservation and lower GC-content. Together, we performed a systematic analysis for histone modification alterations and demonstrated how to identify genomic region-specific elements of epigenetic and genetic regulation and histone modification patterns across different cell types.

Disease Heritability Enrichment of Regulatory Elements Is Concentrated in Elements with Ancient Sequence Age and Conserved Function across Species

Regulatory elements, e.g., enhancers and promoters, have been widely reported to be enriched for disease and complex trait heritability. We investigated how this enrichment varies with the age of the underlying genome sequence, the conservation of regulatory function across species, and the target gene of the regulatory element. We estimated heritability enrichment by applying stratified LD score regression to summary statistics from 41 independent diseases and complex traits (average N = 320K) and meta-analyzing results across traits. Enrichment of human putative enhancers and promoters was larger in elements with older sequence age, assessed via alignment with other species irrespective of conserved functionality: putative enhancer elements with ancient sequence age (older than the split between marsupial and placental mammals) were 8.8× enriched (versus 2.5× for all putative enhancers; p = 3e-14), and promoter elements with ancient sequence age were 13.5× enriched (versus 5.1× for all promoters; p = 5e-16). Enrichment of human putative enhancers and promoters was also larger in elements whose regulatory function was conserved across species, e.g., human putative enhancers that were enhancers in ≥5 of 9 other mammals were 4.6× enriched (p = 5e-12 versus all putative enhancers). Enrichment of human promoters was larger in promoters of loss-of-function intolerant genes: 12.0× enrichment (p = 8e-15 versus all promoters). The mean value of several measures of negative selection within these genomic annotations mirrored all of these findings. Notably, the annotations with these excess heritability enrichments were jointly significant conditional on each other and on our baseline-LD model, which includes a broad set of coding, conserved, regulatory, and LD-related annotations.