Computational methods for identifying enhancer-promoter interactions

Abstract

Background: As parts of the cis-regulatory mechanism of the human genome, interactions between distal enhancers and proximal promoters play a crucial role. Enhancers, promoters, and enhancer-promoter interactions (EPIs) can be detected using many sequencing technologies and computation models. However, a systematic review that summarizes these EPI identification methods and that can help researchers apply and optimize them is still needed.

Results: In this review, we first emphasize the role of EPIs in regulating gene expression and describe a generic framework for predicting enhancer-promoter interaction. Next, we review prediction methods for enhancers, promoters, loops, and enhancer-promoter interactions using different data features that have emerged since 2010, and we summarize the websites available for obtaining enhancers, promoters, and enhancer-promoter interaction datasets. Finally, we review the application of the methods for identifying EPIs in diseases such as cancer.

Conclusions: The advance of computer technology has allowed traditional machine learning, and deep learning methods to be used to predict enhancer, promoter, and EPIs from genetic, genomic, and epigenomic features. In the past decade, models based on deep learning, especially transfer learning, have been proposed for directly predicting enhancer-promoter interactions from DNA sequences, and these models can reduce the parameter training time required of bioinformatics researchers. We believe this review can provide detailed research frameworks for researchers who are beginning to study enhancers, promoters, and their interactions.

Graphical abstract

Keywords

enhancer / promoter / enhancer-promoter interaction / machine learning / deep learning

1 INTRODUCTION

It is known that cis-acting regulatory elements (CREs) are DNA sequences that have transcriptional regulatory functions in the human genome. An enhancer (20- to 400-bp) [1] is a class of non-coding DNA sequences bound by transcription factors [2], and these sequences can interact with short regions of DNA (100–1000 bp), known as promoters, located near the gene transcription start sites (TSS) of a gene [3]. Enhancers and promoters are essential cis-regulatory elements for promoting gene transcription activities over a long distance. The interactions between distal enhancers (even with tens of kilobases) and proximal promoters regulate target genes and inhibit the cis-regulatory mechanism of the human genome [4–9].

Studying the mechanism of enhancer and promoter interactions (EPIs) may help us to understand the regulatory relationships among genes and reveal the genes associated with diseases. Davison et al. showed that EPIs can lead to type I diabetes and multiple sclerosis, and that new genes related to these diseases can be predicted using EPIs [10]. Smemo et al. [11] found the first intron region of the FTO gene in mice and humans, and the homologous gene IRx3 was found to exist in a distal EPI. In the human brain, heart, and lungs high levels of IRx3 gene are expressed; this is very important for controlling weight. Therefore, the study of EPIs, especially cell line-specific EPIs, may provide insight into the mechanisms of gene expression regulation, cell differentiation, and disease. In addition, research on EPIs has provided new methods and ideas for diagnosing and treating disease as well as for developing drugs.

Many sequencing technologies have been developed to generate data and identify enhancer, promoter, and chromosome interactions. For example, epigenomic features such as the histones and transcription factor binding sites (TFBS) data generated by chromatin immunoprecipitation (ChIP-seq) [12,13] and cleavage under targets and release using nuclease (CUT&RUN) [14] technologies have been widely used to identify enhancers and promoters. High-throughput chromosome conformation capture (Hi-C) [15] data (such as BL-Hi-C [16]) is frequently used to call loops (chromosome interactions that connect two distal regulatory elements). Promoter Capture Hi-C [17], Chromatin Interaction Analysis with Paired-End-Tag sequencing (ChIA-PET) [18], and HiChIP [19] can also identify genomic features such as enhancer-promoter interactions. Genetic features such as DNA sequences, pseudo dinucleotide composition (PseDNC), and Pseudo k-tuple nucleotide composition (PseKNC) [20] are also widely used to predict enhancers and promoters. Although the amount of high-throughput sequencing data is increasing rapidly, there are few enhancer-promoter interaction datasets that have been validated by experiments. The prediction of enhancer-promoter interactions using machine learning, deep learning, or other methods is therefore one of the most promising research topics in bioinformatics.

Numerous review articles have been published in recent decades concerning: enhancer interactions, including their role [21] at the genome-wide level; transcription enhancers in animal development, evolution [22], and disease [23]; functional contributions to transcription [24,25]; the functional significance of enhancer chromatin modification [26]; models that describe dynamic three-dimensional chromosome topology related to development enhancers; methods for identifying enhancer target genes [27] and enhancers [28–30]; the mechanisms of EPIs in higher eukaryotes [31]; bioinformatics analysis methods related to EPIs prediction [32–35]; analysis from sequence data [36,37]; and how EPIs control gene expression [38]. However, with the advancement of computational methods in the past decade, research has increasingly proposed methods for detecting enhancer-promoter interaction tools based on traditional machine learning or deep learning, but there has yet to be a global overview of solutions specifically for EPI identification.

In light of this issue, this paper proposes computational models for identifying enhancer-promoter interactions based on high-throughput experimental data published from 2010 to 2022. First, we discuss the relationship between EPIs and gene transcription, and we provide a general framework for enhancer-promoter identification. Next, we discuss in detail recognition methods that have been developed in the last decade for enhancers and promoters, chromatin loops, and enhancer-promoter interactions; we summarize available enhancer and promoter resources, and suggest realistic guidelines for their use. Finally, we review the application of methods for identifying EPIs in diseases such as cancer.

2 REGULATION OF GENE EXPRESSION VIA EPIs

Previous studies [39–41] have shown that the intrachromosomal and interchromosomal communications between enhancer and promoter regulate gene transcription. Transcription from target promoters can be activated by enhancers in interchromatin or intrachromatin over a short distance or a long distance (more than 100 kb) [1] (Fig.1, B), and one enhancer may interact with multiple promoters (Fig.1). He et al. [42] observed that the number of targets for each promoter is 2.92 on average. Some transcription factors may also mediate the interchromosomal interaction between enhancer and promoter (Fig.1). For example, Patel et al. [43] found a T-cell-specific cis-regulatory element in chromosome 16 (TIL16) that can interact with the TAL1 promoter through interchromosomal interaction, and c-Maf and p300 may cooperate to mediate the interchromosomal loop for abnormal activation of TAL1 in T-ALL cells. Therefore, the prediction of enhancers, promoters, and their interactions is vital to our understanding of gene transcription mechanisms.

EPIs can be identified by formulating the problem as follows: “Given two DNA sequences (A and B) described by different data types, first, determine if either A or B can function as an enhancer or a promoter, then determine if A and B are a chromatin loop”. A general process for identifying EPIs is shown in Fig.2, which shows that the identification of EPIs can be divided into four categories:

(i) Given two DNA sequences with transcription factors (TFs), histone marks (provided by ChIP-seq), and chromatin interactions (provided by Hi-C) information, we first need to determine whether the given two DNA sequences are enhancers or promoters by calling peaks, or methods based on traditional machine learning, deep-learning. Then, we need to call loops from Hi-C data to determine whether the two DNA sequences form a chromatin loop.

(ii) Given two DNA sequences with protein chromatin interaction information (provided by ChIA-PET, HiChIP, or PCHi-C), we can call chromatin loops to determine whether the two DNA sequences have EPIs.

(iii) Given two DNA sequences with TFs, histone marks features, and other epigenomic features, two DNA sequences can be identified as EPIs or not by machine-learning-based methods.

(iv) Given two DNA sequences without other information, the two DNA sequences can be identified as EPIs or not by deep-learning-based methods.

Thus we see that the data analysis process can be categorized into the prediction of enhancers, promoters, and EPIs. In the following sections, we describe the prediction of enhancers and promoters, and the identification of EPIs, separately.

3 PREDICTION OF ENHANCER AND PROMOTER

As Tab.1, Tab.2, and Fig.3 show, we can choose methods based on traditional machine learning or deep learning to check if a given DNA sequence is an enhancer or a promoter. To do this, we first need to process the DNA sequence, generate a training set with labels (promoter, enhancer, or none), and then identify enhancers or promoters by traditional machine learning or deep learning.

3.1 Vector representations of DNA sequences

To generate the DNA sequence vectors that can be recognized by traditional machine learning or deep learning, we first need to code the DNA sequence (e.g., ATCGGC…) in one of the following ways. (i) One-hot encoding, although has two problems: (1) the curse of dimensionality and (2) the distance between any pair of one-hot vectors is equal. (ii) To overcome the two problems of one-hot encoding, we use a word embedding algorithm, such as Word2vec [114] or Glove [115], to encode the DNA sequence. For example, dna2vec [116] first transforms a sequence into k-mers (a DNA sequence of length k) and then transforms the k-mers into vectors using Word2vec.

3.1.1 The training sets for enhancers and promoters

There are two ways to obtain enhancer and promoter training sets: (i) Download data sets from a public data repository. For example, we can download the human and mouse enhancer data sets from the SEDB [117] database and can download the eukaryotic promoter from the EPD [118] database. More databases for enhancers and promoters are listed in Tab.3. (ii) Available research has shown that H3K4me1 and H3K27ac enrichment occurs in both enhancers and promoters and that H3K4me1 together with H3K27ac, and a lack of H3K4me3 at the same genomic site can distinguish enhancers from promoters [49]. Additionally, enhancers are enriched with TFBS, Med1. Therefore, we can identify enhancers and promoters by calling peaks from TFBS, H3K27ac, H3K4me1, H3K4me3, or Med1 ChIP-seq data. As Fig.3 shows, we downloaded H3K27ac, H3K4me3, and H3K4me1 ChIP-seq data in the Hela-S3 cell line from the ENCODE platform under accession number ENCSR000AOC, ENCSR000AOF, and ENCSR000APW, respectively. The genome sites with H3K27ac, H3K4me3, and H3K4me1 ChIP-seq signals were identified as promoters. The genome sites with H3K27ac, H3K4me1 ChIP-seq signals, but without H3K4me3 signals were identified as enhancers.

3.1.2 Methods for identifying enhancer/promoter based on traditional machine learning

In machine learning-based methods, the enhancer/promoter identification problem can be reformulated into a binary classification problem (yes or no). Since 2010, support vector machine (SVM) [20,44,46,48,49,51,53,55,59,64,66,86,91,92,140,141], regression [45,60], random forest [47,58,63,65,66,101], boost-based [50,66], and other traditional machine learning methods [52,56,61,62,83,84,87,93–96,103] have all been applied to predict enhancers and promoters. The SVM-based method combined with feature selection has been the most used, even within the last three years. For example, the kmer-SVM [44] first finds the motif related to enhancers by k-mer analysis, then inputs the motif into the SVM model to get the classification results. piEnPred [20] takes advantage of feature extraction techniques such as k-mer, composition of k-spaced nucleic acid pairs (CKSNAP), Dinucleotide-based cross covariance (DCC), PseDNC, and PseKNC to extract features and SVM to classify enhancers and promoters.

Generally, there are three steps to traditional machine learning-based methods. (i) Use of feature extraction techniques to extract features [20,54,57,60,63,84,91,93,96], such as gene expression, histone modification marks, DNA sequence features, and TFs motifs. (ii) Classification of enhancers and promoters by classification algorithms, such as SVM, random forest, or regression. (iii) Tuning of the model parameters and optimization of the target functions using optimization algorithms, such as genetic algorithms [46].

After surveying the accession and citation numbers of these traditional machine-learning methods (Tab.1), we recommend that users who do not want to run code using the web server iEnhancer-2L [53] should identify enhancers and their strengths using pseudo k-tuple nucleotide composition. For users who want to run code by themselves, we recommend that they choose gkm-svm [48], REPTILE [58], and CCS [65]. These tools provide detailed information and example data for users to get up to speed and run them quickly.

3.1.3 Methods for identifying enhancer/promoter based on deep-learning

Methods based on deep-learning primarily focus on training a neural network with DNA sequences or DNA sequences with epigenomic characteristics (such as histone modifications, chromatin accessibility, DNA methylation, or CpG islands) as inputs. Though some scholars have trained their networks with epigenome features [67,68,71,74,75,82], most have done so with only DNA sequences as inputs [69,70,72,73,77–81,85,88–90,98–100,102,104–111,142]. Predicting enhancers and promoters directly from DNA sequences is believed to be more applicable than identifying them from multiple epigenomic features because the epigenomic characteristics data carries with it substantial sequencing costs, and a high rate of false positives. However, prediction methods that use epigenomic characteristics in their inputs are more accurate than those that only use DNA sequences.

Methods based on deep-learning can be roughly divided into the following two steps. (i) Encoding a DNA sequence as in Section “Vector representations of DNA sequence”. (ii) Constructing a neural network to predict the presence of enhancers or promoters, such as CNN [69,73,76,78,81,85,88,98–100,102,104,111], transfer learning [110], or LSTM [82,88,108]. To establish the right characteristics and increase the accuracy of identifying an enhancer or promoter, the above methods either improve the input layer of DNA feature vector representation (for example, dna2vec) or neural network architectures or change the activation functions. Tab.1 and Tab.2 list the available deep-learning-based methods for detecting enhancers and promoters. CSI-ANN [67] was the first deep learning-based method for the identification of enhancers, though Yang et al. [78] have since proposed iEnhancer-GAN to identify enhancers using word embedding, generative adversarial net, and CNN to capture DNA sequence features.

Although computational methods such as traditional machine learning and deep learning have achieved solid results, some problems still exist. One problem is that such methods typically use gene expression data such as chromatin characteristics and histone modification information as features to train models. When gene expression data are missing, these models cannot predict enhancers. Another problem is that enhancers are species-specific. That is, enhancers are expressed differently by different species, so the current methods have low performance in predicting enhancers across species.

For these deep-learning-based methods, we give some suggestions for tool selection. For users who want to predict using ChIP-seq, RNA-seq data, and other features as inputs, we recommend methods based on the input data requirements. For users who wish to identify enhancers and promoters with only DNA sequences as inputs, the number of citations metric (Tab.1 and Tab.2) shows that BiRen [70] and PromID [90] are used frequently for predicting enhancers and promoters, respectively. Online tools including ES-ARCNN [76], iEnhancer-Deep [81] and iPromoter-2L [87] are easy to use and return the prediction results from these methods quickly.

4 PREDICTION OF ENHANCER-PROMOTER INTERACTION

The task of recognizing EPIs is based on the prediction of enhancers and promoters individually in order to determine if there is an interaction between them, and this is a challenging task. First, multiple promoters can be activated by one enhancer, and multiple enhancers can coordinate to regulate one promoter. Secondly, EPI has tissue-specificity [42]. These features result in poor generalization for current EPI recognition methods. The existing EPI recognition methods are divided into three main types: (i) screen EPIs based on high-throughput sequencing experiments, (ii) methods based on traditional machine-learning, and (iii) methods based on deep-learning.

4.1 Generation of EPIs training sets

In surveying the benchmarking EPI data sets used in 12 EPI identification methods (Tab.4), we found 10 methods used the EPI data sets in GM12878, HUVEC, Hela-S3, IMR90, K562, and NHEK cell line proposed by TargetFinder [143]. TargetFinder integrates TFs, histone markers, Dnase-seq, gene expression, and DNA methylation data to predict EPIs. However, before training any model, the EPI data sets need to be augmented, such as with the synthetic minority oversampling technique [156], because of the low ratio of positive to negative data sets (1/35). There are two ways to generate an acceptable EPI dataset.

(i) We can label the active enhancer and promoter regions using ChIP-seq data or annotation files and then annotate chromosome interactions from Hi-C data. For example, EPIP [154] obtained the enhancer data sets and identified the promoter data sets from transcription start site (TSS) annotation files by considering the genomic regions between the 1000 bases upstream and 100 bases downstream of the TSS regions. We can also obtain enhancer and promoter data sets from databases listed in Tab.3. To train an EPI identification model, we can divide the training dataset into positive and negative EPI data sets by overlapping the training data set with the regions of the loops called from Hi-C data [15]. For example, EPIP [154] states that if an enhancer and a promoter overlap with a pair of regions from loops within 30 reads, this pair of enhancer and promoter is considered a positive EPI. We can then use the loop callers listed in Tab.5 to call loops from Hi-C data, such as HiCCUPS [157], HiGlass [159], cLoops [160], FitHiC2 [161], Mustache [162], and HiC-ACT [164]. As Fig.3 displays, to show how to identify EPIs, we downloaded the Hi-C data from 4dnucleome platform under accession number 4DNESCMX7L58, called loops using Mustache [162], and then annotated these loops as enhancer-promoter interactions or promoter-promoter interactions based on ChIP-seq signals.

(ii) We can also obtain EPI data sets by screening loops from target proteins HiChIP, PLAC-seq, or ChIA-PET data. For example, first, H3K27ac HiChIP data can be used to identify enhancer regions by calling loops. Then, we can screen loops that interact with promoters as EPIs. Many available loop callers have been developed for HiChIP, PLAC-seq, and ChIA-PET data. As Tab.5 shows, tools such as HiC-Pro [158], hichipper [166], MAPS [169], FitHiChIP [167], and HiChIP-Peaks [170] have been developed for HiChIP and PLAC-seq data, and tools like ChIA-PET Tool [171], MICC [173], ChIA-PET2 [175], ChIAPoP [176], ChIA-PIPE [177], and MACPET [178] have been developed for ChIA-PET data. Among these tools, HiC-Pro [158] is a pipeline tool for analyzing Hi-C data that includes data pre-processing and calling loops, and FitHiChIP [167] is a fast and memory-efficient loop caller for identifying significant loops. In addition, ChIA-PET2 [175] identifies loops in raw ChIA-PET sequencing reads of different types.

4.2 Methods for identifying EPIs based on traditional machine-learning

The development of high-throughput sequencing technology has produced a huge amount of genomic information, relating to factors such as histone modification and chromatin accessibility. These factors data make it possible to recognize EPIs based on traditional machine learning methods. The basic idea is to use different high-throughput genomic signals as input features of a traditional machine learning model to predict these interactions through statistical calculations. The TF and RNA polymerase ChIP-seq have been reported to be the factors data that can detect EPIs by analyzing epigenomic signals in enhancers and promoters, including TargetFinder [143], EPIP [154], and the XGBoost-based approach [179]. In recent years, boosting ensemble learning methods (e.g., Adaboost [180], gradient boosting decision tree (GBDT) [181], and XGboost [182]) have been used to predict EPIs by constructing multiple weak classifiers. For example, Yu et al. [179] first generated EPI data sets based on chromatin contact data, annotated histones and binding protein data, and a GTF file, and then extracted epigenomic and sequence features. They then trained the XGBoost-based model by five-fold cross-validation in order to predict EPIs. They [179] showed that XGBoost performed better than other machine learning methods, such as TargetFinder [143], random forest [147,183], GBDT [145], or Adaboost [154].

Methods based on traditional machine learning have the advantage of high accuracy for predicting EPIs. However, they have not been widely used for two reasons. The first is the lack of epigenetic characteristics in many cell lines, and the second is that traditional machine-learning-based methods require researchers to possess professional knowledge of epigenetics and manually connect the interaction characteristics.

4.3 Methods for identifying EPIs based on deep-learning

With the development of deep learning, methods for identifying EPIs based on deep-learninghave been proposed for building different neural network architectures in order to learn from DNA sequences without epigenomic characteristics. As is the case for the deep learning-based methods for predicting enhancers and promoters, the process of predicting EPIs includes three steps: (i) embedding the promoter and enhancer DNA sequences based on one-hot encoding or dna2vec, (ii) extracting the promoter and enhancer sequence features based on CNN, LSTM (long short-term memory), or transformer learning, and (iii) predicting EPIs based on the trained network.

Zhuang et al. [148] used one-hot to encode the DNA sequence of enhancers and promoters, but the data storage needed for one-hot encoding consumes a great deal of computer memory and results in the loss of the association information among DNA sequences. EPIVAN [149] and EPI-Mind [153] use dna2vec to embed k-mer into a 100-dimensional vector and contained more information than was the case for one-hot encoding. Singh et al. [146] proposed SPEID to predict long-range EPIs that combine CNN with LSTM. SPEID [146] first inputs the enhancer and promoter vectors encoded by one-hot into CNN, fuses the high-dimensional features extracted from the enhancer and promoter, inputs the fused features into LSTM, and finally outputs the prediction results through the full connection layer. SEPT [155], EPIsHilbert [152], TransEPI [184], and EPI-Mind [153] used transfer learning to get more cross-cell type data features automatically. With the development of deep learning technology, applying transfer learning to the identification of EPIs can reduce the parameter training necessary for each different cell line.

Lastly, we counted the number of citations for available EPI tools, and found that TargetFinder [143] and IM-PET [42] were the most used EPI tools based on traditional machine-learning methods and that EPIVAN [149] and SPEID [146] were the most used EPI tools based on deep-learning methods. Though the web server EPIXplore [185] has not been cited by any article, we suggest that users who do not want to run code access EPIXplore, because EPIXplore integrates IM-PET [42], EpiTensor [186], TargetFinder [143], JEME [187], and 3DPredictor [188], and provides downstream analysis as well as a visualization module. To explore the role that enhancer-promoter interaction structures play in determining normal and pathogenic cell states, we need to use tools that can identify differential EPIs in a process similar to differential expression analysis. Although there is no way to identify differential EPIs directly, we can combine the identification tools for differential loops and EPIs. For example, Lareau et al. proposed diffloop [189] to identify differential loops from ChIA-PET data and identified 1974 differential EPIs from 2 MCF7 and 2 K562 samples. diffHiC [190], FIND [191], HICcompare [192], multiHiCcompare [193], and Serpentine [194] all identify differential loops from Hi-C data.

5 APPLICATIONS OF METHODS FOR IDENTIFYING EPIs IN DISEASES

Genome-wide association studies (GWAS) have revealed that noncoding regulatory sequences, especially the enhancer regions with strong cell specificity, are associated with disease variations [195,196]. Thus, any of the mutations that appear in enhancer-promoter interactions may cause diseases. Carullo et al. [197] discussed in their review study that two types of mutations may disrupt transcriptional regulation (Fig.4). First, the mutations of transcription factors or chromatin modifiers are found at enhancers. Marsman et al. [198] discussed the fact that the gene expression is regulated by transcription factors during cell development, and gene differentiation is regulated by changing loop conformations. For example, as Fig.4 shows, the kit gene is expressed by transcription factors (e.g., GATA-2) in immature erythrocytes, where the enhancers and kit promoter are linked via these transcription factors. When cells mature, other TFs (e.g., GATA-1) that bind to the downstream element (DE) take the place of the GATA-2 TF. TFs including GATA-1 mediate looping between the kit promoter and DE, leading to the disappearance of the loop between enhancer and promoter and the downregulation of kit. Li et al. [199] also showed that the GATA-2 expression and DNA-binding are important for the cell differential process. Second, the mutations of sequence located in enhancers may lead to the loss or gain of functions. Wang et al. [200] proposed the model APRIL to construct long-range regulatory networks and predict novel disease-associated genes with predicted enhancer-gene interactions as inputs (for example, from JEME [187] or IM-PET [42]). In a study by Rodin et al. [201], whole-genome sequencing was performed on 59 donors with autism spectrum disorder (ASD) and 15 control donors and functional enhancers provided by IM-PET [42] to demonstrate that ASD shows an excess of somatic mutations in neural enhancer sequences. Li et al. [18] suggested there is a possibility that mosaic enhancer mutations are associated with ASD risk. In addition, Fachal et al. [202] applied computational enhancer–promoter correlations (using IM-PET [42] and FANTOM5 [60]) and a Bayesian approach (PAINTOR) that they proposed to finely-map 150 breast cancer risk regions and identify 191 likely target genes.

6 CONCLUSION AND FUTURE PROSPECTIVE

Computational methods for identification of enhancers, promoters, and EPIs are valuable for accelerating gene regulation studies, and this paper has reviewed the most important ones to come along over the past decade. We have proposed a basic framework for identifying EPIs and divided the identification methods of EPIs into the following two categories: (i) screening EPIs from ChIP-seq, Hi-C, HiChIP, ChIA-PET, or other High-throughput sequencing technology and (ii) identifying EPIs from DNA sequences, ChIP-seq, Hi-C, or other epigenome data by methods based on traditional machine learning or deep learning. This review also covered enhancer and promoter databases (Tab.3), as well as methods of identifying enhancers (Tab.1), promoters (Tab.2), chromatin loops (Tab.5), and enhancer-promoter interactions (Tab.6). These tables provide practical guidance for readers in selecting methods by model type or input data type in order to identify EPIs. We believe this review can serve as a foundational resource that allows researchers to apply traditional machine learning and deep learning methods to the prediction of enhancers, promoters, and EPIs in future research. We now summarize some important topics for this future work.

First, the initial step of EPI identification based on traditional machine-learning or deep-learning is to pre-process the DNA sequences using one-hot, k-mer, or dna2vec algorithms. However, these methods do not maintain the spatial proximity of the sequence. Designing a new sequence coding method that can maintain the spatial proximity and sequence features is the next task that we urge the EPI research community to undertake.

Secondly, although traditional machine-learning and deep-learning methods have furthered bioinformatics studies for enhancers, promoters, and EPIs for the past ten years, the precision of traditional machine learning is limited because of the high complexity of the source data, its features, and its limited possible model combinations. With recent increases in computing power, however, deep-learning-based methods for identifying EPIs directly from DNA sequences without other epigenome data features have begun to be developed. Furthermore, the rise of transfer learning has reduced the parameter training time needed by bioinformatics researchers. One model can even be fine-tuned by using transfer learning and then transferred to other models for training, which can significantly reduce the amount of needed calculations. For example, transfer learning can be used to predict EPIs [152,153,155,184] across cell lines. An appropriate model trained in one cell line can then be used to predict EPIs directly in another cell line, and this is something that we believe should become a research priority in the future.

Thirdly, with the development of single cell sequence technology, EPI studies at the single-cell level can help us solve the problem of cell heterogeneity, and analyze the mechanism and relationship between individual cells and the body. To accomplish this, available EPI identification methods need to be optimized to accommodate the sparsity of single-cell sequencing data, such as scATAC-seq, scHi-C.

Fourthly, the application of EPI identification methods to exploring tumor-specific EPIs, the effect of mutations on EPIs, and the relationship between EPI formation and gene expression remains the central problem in EPI research. With the development of CRISPR technologies (CRISPR/Cas9, CRISPRa, CRISPRi) and CRISPR screening (Perturb-seq, CRISPRi-FlowFISH etc.), we are now able to identify EPIs or assess the role of EPIs in specific tumors and gene regulatory systems.

References

[1] Bondarenko,V. A., Liu,Y. V., Jiang,Y. I. Studitsky,V. (2003). Communication over a large distance: enhancers and insulators. Biochem. Cell Biol., 81: 241–251

[2] Plank,J. L. (2014). Enhancer function: mechanistic and genome-wide insights come together. Mol. Cell, 55: 5–14

[3] Haberle,V. (2016). Promoter architectures and developmental gene regulation. Semin. Cell Dev. Biol., 57: 11–23

[4] Harismendy,O., Notani,D., Song,X., Rahim,N. G., Tanasa,B., Heintzman,N., Ren,B., Fu,X. D., Topol,E. J., Rosenfeld,M. G. . (2011). 9p21 DNA variants associated with coronary artery disease impair interferon-γ signalling response. Nature, 470: 264–268

[5] Luo,X., Liu,Y., Dang,D., Hu,T., Hou,Y., Meng,X., Zhang,F., Li,T., Wang,C., Li,M. . (2021). 3D Genome of macaque fetal brain reveals evolutionary innovations during primate corticogenesis. Cell, 184: 723–740.e21

[6] Schmitt,A. D., Hu,M., Jung,I., Xu,Z., Qiu,Y., Tan,C. L., Li,Y., Lin,S., Lin,Y., Barr,C. L. . (2016). A compendium of chromatin contact maps reveals spatially active regions in the human genome. Cell Rep., 17: 2042–2059

[7] Policarpi,C., Crepaldi,L., Brookes,E., Nitarska,J., French,S. M., Coatti,A. (2017). Enhancer sines link pol III to pol II transcription in neurons. Cell Rep., 21: 2879–2894

[8] Mumbach,M. R., Satpathy,A. T., Boyle,E. A., Dai,C., Gowen,B. G., Cho,S. W., Nguyen,M. L., Rubin,A. J., Granja,J. M., Kazane,K. R. . (2017). Enhancer connectome in primary human cells identifies target genes of disease-associated DNA elements. Nat. Genet., 49: 1602–1612

[9] May,D., Blow,M. J., Kaplan,T., McCulley,D. J., Jensen,B. C., Akiyama,J. A., Holt,A., Plajzer-Frick,I., Shoukry,M., Wright,C. . (2011). Large-scale discovery of enhancers from human heart tissue. Nat. Genet., 44: 89–93

[10] Davison,L. J., Wallace,C., Cooper,J. D., Cope,N. F., Wilson,N. K., Smyth,D. J., Howson,J. M., Saleh,N., Al-Jeffery,A., Angus,K. L. . (2012). Long-range DNA looping and gene expression analyses identify DEXI as an autoimmune disease candidate gene. Hum. Mol. Genet., 21: 322–333

[11] Smemo,S., Tena,J. J., Kim,K. Gamazon,E. R., Sakabe,N. J., Aneas,I., Credidio,F. L., Sobreira,D. R., Wasserman,N. F. . (2014). Obesity-associated variants within FTO form long-range functional connections with IRX3. Nature, 507: 371–375

[12] Schmidl,C., Rendeiro,A. F., Sheffield,N. C. (2015). ChIPmentation: fast, robust, low-input ChIP-seq for histones and transcription factors. Nat. Methods, 12: 963–965

[13] Carey,M. F., Peterson,C. L. Smale,S. (2009). Chromatin immunoprecipitation (ChIP). Cold Spring Harb. Protoc., 2009: pdb.prot5279

[14] Skene,P. J. (2017). An efficient targeted nuclease strategy for high-resolution mapping of DNA binding sites. eLife, 6: e21856

[15] Belton,J. M., McCord,R. P., Gibcus,J. H., Naumova,N., Zhan,Y. (2012). Hi-C: a comprehensive technique to capture the conformation of genomes. Methods, 58: 268–276

[16] Liang,Z., Li,G., Wang,Z., Djekidel,M. N., Li,Y., Qian,M. Zhang,M. Q. (2017). BL-Hi-C is an efficient and sensitive approach for capturing structural and regulatory chromatin interactions. Nat. Commun., 8: 1622

[17] Schoenfelder,S., Javierre,B. M., Furlan-Magaril,M., Wingett,S. W. (2018). Promoter capture Hi-C: high-resolution, genome-wide profiling of promoter interactions. J. Vis. Exp., (136): e57320

[18] Li,G., Ruan,X., Auerbach,R. K., Sandhu,K. S., Zheng,M., Wang,P., Poh,H. M., Goh,Y., Lim,J., Zhang,J. . (2012). Extensive promoter-centered chromatin interactions provide a topological basis for transcription regulation. Cell, 148: 84–98

[19] Mumbach,M. R., Rubin,A. J., Flynn,R. A., Dai,C., Khavari,P. A., Greenleaf,W. J. Chang,H. (2016). HiChIP: efficient and sensitive analysis of protein-directed genome architecture. Nat. Methods, 13: 919–922

[20] Khan,Z. U., Pi,D. C., Yao,S. L., Nawaz,A., Ali,F. (2021). Pienpred: a bi-layered discriminative model for enhancers and their subtypes via novel cascade multi-level subset feature selection algorithm. Front. Comput. Sci., 15: 1–11

[21] Visel,A., Rubin,E. M. Pennacchio,L. (2009). Genomic views of distant-acting enhancers. Nature, 461: 199–205

[22] Levine,M. (2010). Transcriptional enhancers in animal development and evolution. Curr. Biol., 20: R754–R763

[23] Bulger,M. (2011). Functional and mechanistic diversity of distal transcription enhancers. Cell, 144: 327–339

[24] Kim,T. K. (2015). Architectural and functional commonalities between enhancers and promoters. Cell, 162: 948–959

[25] Ong,C. T. Corces,V. (2011). Enhancer function: new insights into the regulation of tissue-specific gene expression. Nat. Rev. Genet., 12: 283–293

[26] Calo,E. (2013). Modification of enhancer chromatin: what, how, and why? Mol. Cell, 49: 825–837

[27] Yao,L., Berman,B. P. Farnham,P. (2015). Demystifying the secret mission of enhancers: linking distal regulatory elements to target genes. Crit. Rev. Biochem. Mol. Biol., 50: 550–573

[28] Kleftogiannis,D., Kalnis,P. Bajic,V. (2016). Progress and challenges in bioinformatics approaches for enhancer identification. Brief. Bioinform., 17: 967–979

[29] Lim,L. W. K., Chung,H. H., Chong,Y. L. Lee,N. (2018). A survey of recently emerged genome-wide computational enhancer predictor tools. Comput. Biol. Chem., 74: 132–141

[30] Kaur,A., Chauhan,A. P. S., Aggarwal,A. K. (2019). Machine learning based comparative analysis of methods for enhancer prediction in genomic data. In: 2019 2nd International Conference on Intelligent Communication and Computational Techniques (ICCT), 142–145

[31] Kyrchanova,O. (2021). Mechanisms of enhancer-promoter interactions in higher eukaryotes. Int. J. Mol. Sci., 22: 671

[32] Mora,A., Sandve,G. K., Gabrielsen,O. S. (2016). In the loop: promoter-enhancer interactions and bioinformatics. Brief. Bioinform., 17: 980–995

[33] Vanhaeren,T., Divina,F., a-Torres,M., mez-Vela,F., Vanhoof,W. (2020). A comparative study of supervised machine learning algorithms for the prediction of long-range chromatin interactions. Genes (Basel), 11: 985

[34] Tao,H., Li,H., Xu,K., Hong,H., Jiang,S., Du,G., Wang,J., Sun,Y., Huang,X., Ding,Y. . (2021). Computational methods for the prediction of chromatin interaction and organization using sequence and epigenomic profiles. Brief. Bioinform., 22: bbaa405

[35] Xu,H., Zhang,S., Yi,X., Plewczynski,D. Li,M. (2020). Exploring 3D chromatin contacts in gene regulation: the evolution of approaches for the identification of functional enhancer-promoter interaction. Comput. Struct. Biotechnol. J., 18: 558–570

[36] He,C., Li,G., Nadhir,D. M., Chen,Y., Wang,X. Zhang,M. (2016). Advances in computational CHiA-PET data analysis. Quant. Biol., 4: 217–225

[37] Min,X., Lu,F. (2021). Sequence-based deep learning frameworks on enhancer-promoter interactions prediction. Curr. Pharm. Des., 27: 1847–1855

[38] Schoenfelder,S. (2019). Long-range enhancer-promoter contacts in gene expression control. Nat. Rev. Genet., 20: 437–455

[39] Kulaeva,O. I., Nizovtseva,E. V., Polikanov,Y. S., Ulianov,S. V. Studitsky,V. (2012). Distant activation of transcription: mechanisms of enhancer action. Mol. Cell. Biol., 32: 4892–4897

[40] Williams,A., Spilianakis,C. G. Flavell,R. (2010). Interchromosomal association and gene regulation in trans. Trends Genet., 26: 188–197

[41] Maass,P. G., Barutcu,A. R. Rinn,J. (2019). Interchromosomal interactions: a genomic love story of kissing chromosomes. J. Cell Biol., 218: 27–38

[42] He,B., Chen,C., Teng,L. (2014). Global view of enhancer-promoter interactome in human cells. Proc. Natl. Acad. Sci. USA, 111: E2191–E2199

[43] Patel,B., Kang,Y., Cui,K., Litt,M., Riberio,M. S., Deng,C., Salz,T., Casada,S., Fu,X., Qiu,Y. . (2014). Aberrant TAL1 activation is mediated by an interchromosomal interaction in human T-cell acute lymphoblastic leukemia. Leukemia, 28: 349–361

[44] Lee,D., Karchin,R. Beer,M. (2011). Discriminative prediction of mammalian enhancers from DNA sequence. Genome Res., 21: 2167–2180

[45] Taher,L., Narlikar,L. (2012). Clare: cracking the language of regulatory elements. Bioinformatics, 28: 581–583

[46] ndez,M. (2012). Genome-wide enhancer prediction from epigenetic signatures using genetic algorithm-optimized support vector machines. Nucleic Acids Res., 40: e77–e77

[47] Rajagopal,N., Xie,W., Li,Y., Wagner,U., Wang,W., Stamatoyannopoulos,J., Ernst,J., Kellis,M. (2013). RFECS: a random-forest based algorithm for enhancer identification from chromatin state. PLOS Comput. Biol., 9: e1002968

[48] Ghandi,M., Lee,D., Mohammad-Noori,M. Beer,M. (2014). Enhanced regulatory sequence prediction using gapped k-mer features. PLOS Comput. Biol., 10: e1003711

[49] Erwin,G. D., Oksenberg,N., Truty,R. M., Kostka,D., Murphy,K. K., Ahituv,N., Pollard,K. S. Capra,J. (2014). Integrating diverse datasets improves developmental enhancer prediction. PLOS Comput. Biol., 10: e1003677

[50] Lu,Y., Qu,W., Shan,G. (2015). Delta: a distal enhancer locating tool based on AdaBoost algorithm and shape features of chromatin modifications. PLoS One, 10: e0130622

[51] Kleftogiannis,D., Kalnis,P. Bajic,V. (2015). DEEP: a general computational framework for predicting enhancers. Nucleic Acids Res., 43: e6

[52] Liu,B. (2016). Ienhancer-psedeknc: identification of enhancers and their subgroups based on pseudo degenerate kmer nucleotide composition. Neurocomputing, 217: 46–52

[53] Liu,B., Fang,L., Long,R., Lan,X. Chou,K. (2016). iEnhancer-2L: a two-layer predictor for identifying enhancers and their strength by pseudo k-tuple nucleotide composition. Bioinformatics, 32: 362–369

[54] Jia,C. (2016). EnhancerPred: a predictor for discovering enhancers based on the combination and selection of multiple features. Sci. Rep., 6: 38741

[55] Xu,J., Hu,H. (2016). Lmethyr-svm: predict human enhancers using low methylated regions based on weighted support vector machines. PLoS One, 11: e0163491

[56] Colbran,L. L., Chen,L. Capra,J. (2017). Short DNA sequence patterns accurately identify broadly active human enhancers. BMC Genomics, 18: 536

[57] He,W. (2017). EnhancerPred2.0: predicting enhancers and their strength based on position-specific trinucleotide propensity and electron-ion interaction potential feature selection. Mol. Biosyst., 13: 767–774

[58] He,Y., Gorkin,D. U., Dickel,D. E., Nery,J. R., Castanon,R. G., Lee,A. Y., Shen,Y., Visel,A., Pennacchio,L. A., Ren,B. . (2017). Improved regulatory element prediction based on tissue-specific local epigenomic signatures. Proc. Natl. Acad. Sci. USA, 114: E1633–E1640

[59] Liu,B., Li,K., Huang,D. S. Chou,K. (2018). iEnhancer-EL: identifying enhancers and their strength with ensemble learning approach. Bioinformatics, 34: 3835–3842

[60] Kleftogiannis,D., Ashoor,H. Bajic,V. (2018). Tels: a novel computational framework for identifying motif signatures of transcribed enhancers. Genom. Proteom. Bioinf., 16: 332–341

[61] Singh,A. P., Mishra,S. (2018). Sequence based prediction of enhancer regions from DNA random walk. Sci. Rep., 8: 15912

[62] Sethi,A., Gu,M., Gumusgoz,E., Chan,L., Yan,K. K., Rozowsky,J., Barozzi,I., Afzal,V., Akiyama,J. A., Plajzer-Frick,I. . (2020). Supervised enhancer prediction with epigenetic pattern recognition and targeted validation. Nat. Methods, 17: 807–814

[63] Lim,D. Y., Khanal,J., Tayara,H. Chong,K. (2021). Ienhancer-rf: identifying enhancers and their strength by enhanced feature representation using random forest. Chemom. Intell. Lab. Syst., 212: 104284

[64] Lyu,Y., Zhang,Z., Li,J., He,W., Ding,Y. (2021). Ienhancer-kl: a novel two-layer predictor for identifying enhancers by position specific of nucleotide composition. IEEE/ACM Trans. Comput. Biol. Bioinformatics, 18: 2809–2815

[65] Niu,X., Deng,K., Liu,L., Yang,K. (2021). A statistical framework for predicting critical regions of p53-dependent enhancers. Brief. Bioinform., 22: bbaa053

[66] Basith,S., Hasan,M. M., Lee,G., Wei,L. (2021). Integrative machine learning framework for the identification of cell-specific enhancers from the human genome. Brief. Bioinform., 22: bbab252

[67] Firpi,H. A., Ucar,D. (2010). Discover regulatory DNA elements using chromatin signatures and artificial neural network. Bioinformatics, 26: 1579–1586

[68] Liu,F., Li,H., Ren,C., Bo,X. (2016). PEDLA: predicting enhancers with a deep learning-based algorithmic framework. Sci. Rep., 6: 28517

[69] Min,X., Zeng,W., Chen,S., Chen,N., Chen,T. (2017). Predicting enhancers with deep convolutional neural networks. BMC Bioinformatics, 18: 478

[70] Yang,B., Liu,F., Ren,C., Ouyang,Z., Xie,Z., Bo,X. (2017). BiRen: predicting enhancers with a deep-learning-based model using the DNA sequence alone. Bioinformatics, 33: 1930–1936

[71] Thibodeau,A., Uyar,A., Khetan,S., Stitzel,M. L. (2018). A neural network based model effectively predicts enhancers from clinical ATAC-seq samples. Sci. Rep., 8: 16048

[72] Le,N. Q. K., Yapp,E. K. Y., Ho,Q. T., Nagasundaram,N., Ou,Y. Y. Yeh,H. (2019). iEnhancer-5Step: identifying enhancers using hidden information of DNA sequences via Chou’s 5-step rule and word embedding. Anal. Biochem., 571: 53–61

[73] Khanal,J., Tayara,H. Chong,K. (2020). Identifying enhancers and their strength by the integration of word embedding and convolution neural network. IEEE Access, 8: 58369–58376

[74] Chen,S., Gan,M., Lv,H. (2021). Deepcape: a deep convolutional neural network for the accurate prediction of enhancers. Genom. Proteom. Bioinf., 19: 565–577

[75] Chen,Z., Zhang,J., Liu,J., Dai,Y., Lee,D., Min,M. R., Xu,M. (2021). DECODE: a Deep-learning framework for condensing enhancers and refining boundaries with large-scale functional assays. Bioinformatics, 37: i280–i288

[76] Zhang,T. H., Flores,M. (2021). ES-ARCNN: predicting enhancer strength by using data augmentation and residual convolutional neural network. Anal. Biochem., 618: 114120

[77] Inayat,N., Khan,M., Iqbal,N., Khan,S., Raza,M., Khan,D. M., Khan,A. Wei,D. (2021). Ienhancer-dhf: identification of enhancers and their strengths using optimize deep neural network with multiple features extraction methods. IEEE Access, 9: 40783–40796

[78] Yang,R., Wu,F., Zhang,C. (2021). Ienhancer-gan: a deep learning framework in combination with word embedding and sequence generative adversarial net to identify enhancers and their strength. Int. J. Mol. Sci., 22: 3589

[79] Gao,Y., Chen,Y., Feng,H., Zhang,Y. (2022). Ricenn: prediction of rice enhancers with neural network based on DNA sequences. Interdiscip. Sci., 14: 555–565

[80] Amilpur,S. (2022). A sequence-based two-layer predictor for identifying enhancers and their strength through enhanced feature extraction. J. Bioinform. Comput. Biol., 20: 2250005

[81] Kamran,H., Tahir,M., Tayara,H. Chong,K. (2022). Ienhancer-deep: a computational predictor for enhancer sites and their strength using deep learning. Appl. Sci. (Basel), 12: 2120

[82] Zhao,S., Pan,Q., Zou,Q., Ju,Y., Shi,L. (2022). Identifying and classifying enhancers by dinucleotide-based auto-cross covariance and attention-based Bi-LSTM. Comput. Math. Methods Med., 2022: 7518779

[83] Song,K. (2012). Recognition of prokaryotic promoters based on a novel variable-window Z-curve method. Nucleic Acids Res., 40: 963–971

[84] Li,Y., Chen,C. Y. Wasserman,W. (2016). Deep feature selection: theory and application to identify enhancers and promoters. J. Comput. Biol., 23: 322–336

[85] Umarov,R. K. Solovyev,V. (2017). Recognition of prokaryotic and eukaryotic promoters using convolutional deep learning neural networks. PLoS One, 12: e0171410

[86] Coelho,R. V., de Avila E Silva,S., Echeverrigaray,S., Delamare,A. P. L. Delamare,A. P. (2018). Bacillus subtilis promoter sequences data set for promoter prediction in Gram-positive bacteria. Data Brief, 19: 264–270

[87] Liu,B., Yang,F., Huang,D. S. Chou,K. (2018). iPromoter-2L: a two-layer predictor for identifying promoters and their types by multi-window-based PseKNC. Bioinformatics, 34: 33–40

[88] Oubounyt,M., Louadi,Z., Tayara,H. Chong,K. (2019). Deepromoter: robust promoter predictor using deep learning. Front. Genet., 10: 286

[89] Le,N. Q. K., Yapp,E. K. Y., Nagasundaram,N. Yeh,H. (2019). Classifying promoters by interpreting the hidden information of DNA sequences via deep learning and combination of continuous fasttext N-grams. Front. Bioeng. Biotechnol., 7: 305

[90] Umarov,R., Kuwahara,H., Li,Y., Gao,X. (2019). Promoter analysis and prediction in the human genome using sequence-based deep learning models. Bioinformatics, 35: 2730–2737

[91] Lai,H. Y., Zhang,Z. Y., Su,Z. D., Su,W., Ding,H., Chen,W. (2019). Iproep: a computational predictor for predicting promoter. Mol. Ther. Nucleic Acids, 17: 337–346

[92] Liu,B. (2019). Ipromoter-2l2. 0: identifying promoters and their types by combining smoothing cutting window algorithm and sequence-based features. Mol. Ther. Nucleic Acids, 18: 80–87

[93] Rahman,M. S., Aktar,U., Jani,M. R. (2019). iPromoter-FSEn: identification of bacterial σ⁷⁰ promoter sequences using feature subspace based ensemble classifier. Genomics, 111: 1160–1166

[94] Rahman,M. S., Aktar,U., Jani,M. R. (2019). iPro70-FMWin: identifying Sigma70 promoters using multiple windowing and minimal features. Mol. Genet. Genomics, 294: 69–84

[95] Xiao,X., Xu,Z. C., Qiu,W. R., Wang,P., Ge,H. T. Chou,K. (2019). iPSW(2L)-PseKNC: a two-layer predictor for identifying promoters and their strength by hybrid features via pseudo k-tuple nucleotide composition. Genomics, 111: 1785–1793

[96] Zhang,M., Li,F., Marquez-Lago,T. T., Leier,A., Fan,C., Kwoh,C. K., Chou,K. C., Song,J. (2019). MULTiPly: a novel multi-layer predictor for discovering general and specific types of promoters. Bioinformatics, 35: 2957–2965

[97] Chen,Y. L., Guo,D. H. Li,Q. (2020). An energy model for recognizing the prokaryotic promoters based on molecular structure. Genomics, 112: 2072–2079

[98] Amin,R., Rahman,C. R., Ahmed,S., Sifat,M. H. R., Liton,M. N. K., Rahman,M. M., Khan,M. Z. H. (2020). iPromoter-BnCNN: a novel branched CNN-based predictor for identifying and classifying sigma promoters. Bioinformatics, 36: 4869–4875

[99] Tayara,H., Tahir,M. Chong,K. (2020). Identification of prokaryotic promoters and their strength by integrating heterogeneous features. Genomics, 112: 1396–1403

[100] Shujaat,M., Wahab,A., Tayara,H. Chong,K. (2020). Pcpromoter-CNN: a CNN-based prediction and classification of promoters. Genes (Basel), 11: 1529

[101] Liang,Y., Zhang,S., Qiao,H. (2021). iPromoter-ET: identifying promoters and their strength by extremely randomized trees-based feature selection. Anal. Biochem., 630: 114335

[102] Shujaat,M., Lee,S. B., Tayara,H. Chong,K. (2021). Cr-prom: a convolutional neural network-based model for the prediction of rice promoters. IEEE Access, 9: 81485–81491

[103] Lyu,Y., He,W., Li,S., Zou,Q. (2021). Ipro2l-pstknc: a two-layer predictor for discovering various types of promoters by position specific of nucleotide composition. IEEE J. Biomed. Health Inform., 25: 2329–2337

[104] Sun,A., Xiao,X. (2021). Iptt(2 L)-CNN: a two-layer predictor for identifying promoters and their types in plant genomes by convolutional neural network. Comput. Math. Methods Med., 2021: 6636350

[105] Zhu,Y., Li,F., Xiang,D., Akutsu,T., Song,J. (2021). Computational identification of eukaryotic promoters based on cascaded deep capsule neural networks. Brief. Bioinform., 22: bbaa299

[106] Bhukya,R., Kumari,A., Amilpur,S. Dasari,C. (2022). PPred-PCKSM: a multi-layer predictor for identifying promoter and its variants using position based features. Comput. Biol. Chem., 97: 107623

[107] Li,H., Shi,L., Gao,W., Zhang,Z., Zhang,L., Zhao,Y. (2022). dPromoter-XGBoost: detecting promoters and strength by combining multiple descriptors and feature selection using XGBoost. Methods, 204: 215–222

[108] Li,Q. W., Zhang,L. C., Xu,L., Zou,Q., Wu,J. Li,Q. (2022). Identification and classification of promoters using the attention mechanism based on long short-term memory. Front. Comput. Sci., 16: 164348

[109] Qiao,H., Zhang,S., Xue,T., Wang,J. (2022). iPro-GAN: a novel model based on generative adversarial learning for identifying promoters and their strength. Comput. Methods Programs Biomed., 215: 106625

[110] Wang,Y., Peng,Q., Mou,X., Wang,X., Li,H., Han,T., Sun,Z. (2022). A successful hybrid deep learning model aiming at promoter identification. BMC Bioinformatics, 23: 206

[111] Wei,P. J., Pang,Z. Z., Jiang,L. J., Tan,D. Y., Su,Y. S. Zheng,C. (2022). Promoter prediction in nannochloropsis based on densely connected convolutional neural networks. Methods, 204: 38–46

[112] Mifsud,B., Tavares-Cadete,F., Young,A. N., Sugar,R., Schoenfelder,S., Ferreira,L., Wingett,S. W., Andrews,S., Grey,W., Ewels,P. A. . (2015). Mapping long-range promoter contacts in human cells with high-resolution capture Hi-C. Nat. Genet., 47: 598–606

[113] Javierre,B. M., Burren,O. S., Wilder,S. P., Kreuzhuber,R., Hill,S. M., Sewitz,S., Cairns,J., Wingett,S. W., rnai,C., Thiecke,M. J. . (2016). Lineage-specific genome architecture links enhancers and non-coding disease variants to target gene promoters. Cell, 167: 1369–1384.e19

[114] Mikolov,T., Yih,W. (2013). Linguistic regularities in continuous space word representations. In: Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human language technologies, 746–751

[115] Pennington,J., Socher,R. Manning,C. (2014). Glove: global vectors for word representation. In: Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), 1532–1543

[116] Ng,P. (2017). Dna2vec: consistent vector representations of variable-length k-mers. arXiv, 1701.06279

[117] Jiang,Y., Qian,F., Bai,X., Liu,Y., Wang,Q., Ai,B., Han,X., Shi,S., Zhang,J., Li,X. . (2019). SEdb: a comprehensive human super-enhancer database. Nucleic Acids Res., 47: D235–D243

[118] rier,R. C., Praz,V., Junier,T., Bonnard,C. (2000). The eukaryotic promoter database (EPD). Nucleic Acids Res., 28: 302–303

[119] Ferretti,V., Poitras,C., Bergeron,D., Coulombe,B., Robert,F. (2007). PReMod: a database of genome-wide mammalian cis-regulatory module predictions. Nucleic Acids Res., 35: D122–D126

[120] Andersson,R., Gebhard,C., Miguel-Escalada,I., Hoof,I., Bornholdt,J., Boyd,M., Chen,Y., Zhao,X., Schmidl,C., Suzuki,T. . (2014). An atlas of active enhancers across human cell types and tissues. Nature, 507: 455–461

[121] Visel,A., Minovitsky,S., Dubchak,I. Pennacchio,L. (2007). VISTA Enhancer Browser—a database of tissue-specific human enhancers. Nucleic Acids Res., 35: D88–D92

[122] Hoke,H. A., Lin,C. Y., Lau,A., Orlando,D. A., Vakoc,C. R., Bradner,J. E., Lee,T. I. Young,R. (2013). Selective inhibition of tumor oncogenes by disruption of super-enhancers. Cell, 153: 320–334

[123] Bai,X., Shi,S., Ai,B., Jiang,Y., Liu,Y., Han,X., Xu,M., Pan,Q., Wang,F., Wang,Q. . (2020). ENdb: a manually curated database of experimentally supported enhancers for human and mouse. Nucleic Acids Res., 48: D51–D57

[124] Wei,Y., Zhang,S., Shang,S., Zhang,B., Li,S., Wang,X., Wang,F., Su,J., Wu,Q., Liu,H. . (2016). SEA: a super-enhancer archive. Nucleic Acids Res., 44: D172–D179

[125] Cai,Z. N., Cui,Y., Tan,Z. Y., Zhang,G. H., Tan,Z. Y., Zhang,X. L. Peng,Y. (2019). RAEdb: A database of enhancers identified by high-throughput reporter assays. Database (Oxford), bay140

[126] Guo,Z. W., Xie,C., Li,K., Zhai,X. M., Cai,G. X., Yang,X. X. Wu,Y. (2019). Seler: a database of super-enhancer-associated lncRNA-directed transcriptional regulation in human cancers. Database (Oxford), baz027

[127] Zeng,W. W., Min,X. (2019). Endisease: a manually curated database for enhancer-disease associations. Database (Oxford), baz020

[128] Huang,M., Wang,Y., Yang,M., Yan,J., Yang,H., Zhuang,W., Xu,Y., Koeffler,H. P., Lin,D. C. (2020). dbInDel: a database of enhancer-associated insertion and deletion variants by analysis of H3K27ac ChIP-Seq. Bioinformatics, 36: 1649–1651

[129] Kumar,R., Lathwal,A., Kumar,V., Patiyal,S., Raghav,P. K. Raghava,G. P. (2020). CancerEnD: a database of cancer associated enhancers. Genomics, 112: 3696–3702

[130] Vasyuchenko,E. P., Orekhov,P. S., Armeev,G. A. Bozdaganyan,M. (2021). Cpe-db: an open database of chemical penetration enhancers. Pharmaceutics, 13: 66

[131] Jin,W., Jiang,G., Yang,Y., Yang,J., Yang,W., Wang,D., Niu,X., Zhong,R., Zhang,Z. (2022). Animal-eRNAdb: a comprehensive animal enhancer RNA database. Nucleic Acids Res., 50: D46–D53

[132] Shahmuradov,I. A., Gammerman,A. J., Hancock,J. M., Bramley,P. M. Solovyev,V. (2003). PlantProm: a database of plant promoter sequences. Nucleic Acids Res., 31: 114–117

[133] Smirnova,O. G., Ibragimova,S. S. Kochetov,A. (2012). Simple database to select promoters for plant transgenesis. Transgenic Res., 21: 429–437

[134] Grienberg,I. (2005). Osteo-Promoter Database (OPD)—promoter analysis in skeletal cells. BMC Genomics, 6: 46

[135] Morris,R. T., Connor,T. R. Wyrick,J. (2008). Osiris: an integrated promoter database for Oryza sativa L. Bioinformatics, 24: 2915–2917

[136] Chen,X., Wu,J. M., Hornischer,K., Kel,A. (2006). Tiprod: the tissue-specific promoter database. Nucleic Acids Res., 34: D104–D107

[137] Nishikata,K., Cox,R. S. Shimoyama,S., Yoshida,Y., Matsui,M., Makita,Y. (2014). Database construction for PromoterCAD: synthetic promoter design for mammals and plants. ACS Synth. Biol., 3: 192–196

[138] Dreos,R., Ambrosini,G., rier,R. C. (2015). The Eukaryotic Promoter Database: expansion of EPDnew and new promoter analysis tools. Nucleic Acids Res., 43: D92–D96

[139] Su,W., Liu,M. L., Yang,Y. H., Wang,J. S., Li,S. H., Lv,H., Dao,F. Y., Yang,H. (2021). Ppd: a manually curated database for experimentally verified prokaryotic promoters. J. Mol. Biol., 433: 166860

[140] Gordon,L., Chervonenkis,A. Y., Gammerman,A. J., Shahmuradov,I. A. Solovyev,V. (2003). Sequence alignment kernel for recognition of promoter regions. Bioinformatics, 19: 1964–1971

[141] Towsey,M., Timms,P., Hogan,J. Mathews,S. (2008). The cross-species prediction of bacterial promoters using a support vector machine. Comput. Biol. Chem., 32: 359–366

[142] Knudsen,S. (1999). Promoter2.0: for the recognition of PolII promoter sequences. Bioinformatics, 15: 356–361

[143] Whalen,S., Truty,R. M. Pollard,K. (2016). Enhancer-promoter interactions are encoded by complex genomic signatures on looping chromatin. Nat. Genet., 48: 488–496

[144] Yang,Y., Zhang,R., Singh,S. (2017). Exploiting sequence-based features for predicting enhancer-promoter interactions. Bioinformatics, 33: i252–i260

[145] Zeng,W., Wu,M. (2018). Prediction of enhancer-promoter interactions via natural language processing. BMC Genomics, 19: 84

[146] Singh,S., Yang,Y., czos,B. (2019). Predicting enhancer-promoter interaction from genomic sequence with deep neural networks. Quant. Biol., 7: 122–137

[147] Zhang,T. (2019). An approach for recognition of enhancer-promoter associations based on random forest. In: Proceedings of the 2019 4th International Conference on Biomedical Signal and Image Processing (ICBIP 2019), 46–50

[148] Zhuang,Z., Shen,X. (2019). A simple convolutional neural network for prediction of enhancer-promoter interactions with DNA sequence data. Bioinformatics, 35: 2899–2906

[149] Hong,Z., Zeng,X., Wei,L. (2020). Identifying enhancer-promoter interactions with neural network based on pre-trained DNA vectors and attention mechanism. Bioinformatics, 36: 1037–1043

[150] Singh,A. P., Mishra,S. (2018). Sequence based prediction of enhancer regions from DNA random walk. Sci Rep. 8, 15912

[151] Min,X., Ye,C., Liu,X. (2021). Predicting enhancer-promoter interactions by deep learning and matching heuristic. Brief. Bioinform., 22: bbaa254

[152] Zhang,M., Hu,Y. (2021). Epishilbert: prediction of enhancer-promoter interactions via hilbert curve encoding and transfer learning. Genes (Basel), 12: 1385

[153] Ni,Y., Fan,L., Wang,M., Zhang,N., Zuo,Y. (2022). Epi-mind: identifying enhancer-promoter interactions based on transformer mechanism. Interdiscip. Sci., 14: 786–794

[154] Talukder,A., Saadat,S., Li,X. (2019). EPIP: a novel approach for condition-specific enhancer-promoter interaction prediction. Bioinformatics, 35: 3877–3883

[155] Jing,F., Zhang,S. W. (2020). Prediction of enhancer-promoter interactions using the cross-cell type information and domain adversarial neural network. BMC Bioinformatics, 21: 507

[156] Chawla,N. V., Bowyer,K. W., Hall,L. O. Kegelmeyer,W. (2002). Smote: synthetic minority over-sampling technique. J. Artif. Intell. Res., 16: 321–357

[157] Rao,S. S. P., Huntley,M. H., Durand,N. C., Stamenova,E. K., Bochkov,I. D., Robinson,J. T., Sanborn,A. L., Machol,I., Omer,A. D., Lander,E. S. . (2014). A 3D map of the human genome at kilobase resolution reveals principles of chromatin looping. Cell, 159: 1665–1680

[158] Servant,N., Varoquaux,N., Lajoie,B. R., Viara,E., Chen,C. Vert,J. Heard,E., Dekker,J. (2015). HiC-Pro: an optimized and flexible pipeline for Hi-C data processing. Genome Biol., 16: 259

[159] Kerpedjiev,P., Abdennur,N., Lekschas,F., McCallum,C., Dinkla,K., Strobelt,H., Luber,J. M., Ouellette,S. B., Azhir,A., Kumar,N. . (2018). HiGlass: web-based visual exploration and analysis of genome interaction maps. Genome Biol., 19: 125

[160] Cao,Y., Chen,Z., Chen,X., Ai,D., Chen,G., McDermott,J., Huang,Y., Guo,X. Han,J. (2020). Accurate loop calling for 3D genomic data with cLoops. Bioinformatics, 36: 666–675

[161] Kaul,A., Bhattacharyya,S. (2020). Identifying statistically significant chromatin contacts from Hi-C data with FitHiC2. Nat. Protoc., 15: 991–1012

[162] Roayaei Ardakany,A., Gezer,H. T., Lonardi,S. (2020). Mustache: multi-scale detection of chromatin loops from Hi-C and Micro-C maps using scale-space representation. Genome Biol., 21: 256

[163] Krietenstein,N., Abraham,S., Venev,S. V., Abdennur,N., Gibcus,J., Hsieh,T. S., Parsi,K. M., Yang,L., Maehr,R., Mirny,L. A. . (2020). Ultrastructural details of mammalian chromosome architecture. Mol. Cell, 78: 554–565.e7

[164] Lagler,T. M., Abnousi,A., Hu,M., Yang,Y. (2021). HiC-ACT: improved detection of chromatin interactions from Hi-C data via aggregated Cauchy test. Am. J. Hum. Genet., 108: 257–268

[165] Lee,H. Seo,P. (2021). Hicore: Hi-c analysis for identification of core chromatin looping regions with higher resolution. Mol. Cells, 44: 883–892

[166] Lareau,C. A. Aryee,M. (2018). hichipper: a preprocessing pipeline for calling DNA loops from HiChIP data. Nat. Methods, 15: 155–156

[167] Bhattacharyya,S., Chandra,V., Vijayanand,P. (2019). Identification of significant chromatin contacts from HiChIP data by FitHiChIP. Nat. Commun., 10: 4221

[168] Fang,R., Yu,M., Li,G., Chee,S., Liu,T., Schmitt,A. D. (2016). Mapping of long-range chromatin interactions by proximity ligation-assisted ChIP-seq. Cell Res., 26: 1345–1348

[169] Juric,I., Yu,M., Abnousi,A., Raviram,R., Fang,R., Zhao,Y., Zhang,Y., Qiu,Y., Yang,Y., Li,Y. . (2019). MAPS: model-based analysis of long-range chromatin interactions from PLAC-seq and HiChIP experiments. PLOS Comput. Biol., 15: e1006982

[170] Shi,C., Rattray,M. (2020). HiChIP-Peaks: a HiChIP peak calling algorithm. Bioinformatics, 36: 3625–3631

[171] Li,G., Fullwood,M. J., Xu,H., Mulawadi,F. H., Velkov,S., Vega,V., Ariyaratne,P. N., Mohamed,Y. B., Ooi,H. S., Tennakoon,C. . (2010). ChIA-PET tool for comprehensive chromatin interaction analysis with paired-end tag sequencing. Genome Biol., 11: R22

[172] Paulsen,J., dland,E. A., Holden,L., Holden,M. (2014). A statistical model of ChIA-PET data for accurate detection of chromatin 3D interactions. Nucleic Acids Res., 42: e143

[173] He,C., Zhang,M. Q. (2015). MICC: an R package for identifying chromatin interactions from ChIA-PET data. Bioinformatics, 31: 3832–3834

[174] Djekidel,M. N., Liang,Z., Wang,Q., Hu,Z., Li,G., Chen,Y. Zhang,M. (2015). 3CPET: finding co-factor complexes from ChIA-PET data using a hierarchical Dirichlet process. Genome Biol., 16: 288

[175] Li,G., Chen,Y., Snyder,M. P. Zhang,M. (2017). ChIA-PET2: a versatile and flexible pipeline for ChIA-PET data analysis. Nucleic Acids Res., 45: e4

[176] Huang,W., Medvedovic,M., Zhang,J. (2019). ChIAPoP: a new tool for ChIA-PET data analysis. Nucleic Acids Res., 47: e37

[177] Lee,B., Wang,J., Cai,L., Kim,M., Namburi,S., Tjong,H., Feng,Y., Wang,P., Tang,Z., Abbas,A. . (2020). ChIA-PIPE: a fully automated pipeline for comprehensive ChIA-PET data analysis and visualization. Sci. Adv., 6: eaay2078

[178] Vardaxis,I., Rye,M. B. Lindqvist,B. (2020). MACPET: model-based analysis for ChIA-PET. Biostatistics, 21: 625–639

[179] Yu,X. J., Zhou,J. G., Zhao,M. M., Yi,C., Duan,Q., Zhou,W. (2020). Exploiting XGboost for predicting enhancer-promoter interactions. Curr. Bioinform., 15: 1036–1045

[180] Bartlett,P. (2006). Adaboost is consistent. Adv. Neural Inf. Process. Syst., 8: 2347–2368

[181] Zhang,S., Dong,X. (2012). Synonym recognition based on user behaviors in E-commerce. Journal of Chinese Information Processing (in Chinese), 26: 79–85

[182] Chen,T. (2016). XgBoost: a scalable tree boosting system. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 785–794

[183] Feng,Z. X. Li,Q. (2017). Recognition of long-range enhancer-promoter interactions by adding genomic signatures of segmented regulatory regions. Genomics, 109: 341–352

[184] Chen,K., Zhao,H. (2022). Capturing large genomic contexts for accurately predicting enhancer-promoter interactions. Brief. Bioinform., 23: bbab577

[185] Tang,L., Zhong,Z., Lin,Y., Yang,Y., Wang,J., Martin,J. F. (2022). EPIXplorer: a web server for prediction, analysis and visualization of enhancer-promoter interactions. Nucleic Acids Res., 50: W290–W297

[186] Zhu,Y., Chen,Z., Zhang,K., Wang,M., Medovoy,D., Whitaker,J. W., Ding,B., Li,N., Zheng,L. (2016). Constructing 3D interaction maps from 1D epigenomes. Nat. Commun., 7: 10812

[187] Cao,Q., Anyansi,C., Hu,X., Xu,L., Xiong,L., Tang,W., Mok,M. T. S., Cheng,C., Fan,X., Gerstein,M. . (2017). Reconstruction of enhancer-target networks in 935 samples of human primary cells, tissues and cell lines. Nat. Genet., 49: 1428–1436

[188] Belokopytova,P. S., Nuriddinov,M. A., Mozheiko,E. A., Fishman,D. (2020). Quantitative prediction of enhancer-promoter interactions. Genome Res., 30: 72–84

[189] Lareau,C. A. Aryee,M. (2018). diffloop: a computational framework for identifying and analyzing differential DNA loops from sequencing data. Bioinformatics, 34: 672–674

[190] Lun,A. T. Smyth,G. (2015). diffHic: a Bioconductor package to detect differential genomic interactions in Hi-C data. BMC Bioinformatics, 16: 258

[191] Djekidel,M. N., Chen,Y. Zhang,M. (2018). FIND: difFerential chromatin INteractions Detection using a spatial Poisson process. Genome Res., 28: 412–422

[192] Stansfield,J. C., Cresswell,K. G., Vladimirov,V. I. Dozmorov,M. (2018). HiCcompare: an R-package for joint normalization and comparison of Hi-C datasets. BMC Bioinformatics, 19: 279

[193] Stansfield,J. C., Cresswell,K. G. Dozmorov,M. (2019). multiHiCcompare: joint normalization and comparative analysis of complex Hi-C experiments. Bioinformatics, 35: 2916–2923

[194] Baudry,L., Millot,G. A., Thierry,A., Koszul,R. Scolari,V. (2020). Serpentine: a flexible 2D binning method for differential Hi-C analysis. Bioinformatics, 36: 3645–3651

[195] Karczewski,K. J., Dudley,J. T., Kukurba,K. R., Chen,R., Butte,A. J., Montgomery,S. B. (2013). Systematic functional regulatory assessment of disease-associated variants. Proc. Natl. Acad. Sci. USA, 110: 9607–9612

[196] Corradin,O., Saiakhova,A., Akhtar-Zaidi,B., Myeroff,L., Willis,J., Cowper-Sal lari,R., Lupien,M., Markowitz,S. Scacheri,P. (2014). Combinatorial effects of multiple enhancer variants in linkage disequilibrium dictate levels of gene expression to confer susceptibility to common traits. Genome Res., 24: 1–13

[197] Carullo,N. V. N. Day,J. (2019). Genomic enhancers in brain health and disease. Genes (Basel), 10: 43

[198] Marsman,J. Horsfield,J. (2012). Long distance relationships: enhancer-promoter communication and dynamic gene transcription. Gene Regulatory Mechanisms., 1819: 1217–1227

[199] Li,Y., He,Y., Liang,Z., Wang,Y., Chen,F., Djekidel,M. N., Li,G., Zhang,X., Xiang,S., Wang,Z. . (2018). Alterations of specific chromatin conformation affect ATRA-induced leukemia cell differentiation. Cell Death Dis., 9: 200

[200] Wang,H., Yang,J., Zhang,Y. (2021). Discover novel disease-associated genes based on regulatory networks of long-range chromatin interactions. Methods, 189: 22–33

[201] Rodin,R. E., Dou,Y., Kwon,M., Sherman,M. A., Gama,A. M., Doan,R. N., Rento,L. M., Girskis,K. M., Bohrson,C. L., Kim,S. N. . (2021). The landscape of somatic mutation in cerebral cortex of autistic and neurotypical individuals revealed by ultra-deep whole-genome sequencing. Nat. Neurosci., 24: 176–185

[202] Fachal,L., Aschard,H., Beesley,J., Barnes,D. R., Allen,J., Kar,S., Pooley,K. A., Dennis,J., Michailidou,K., Turman,C. . (2020). Fine-mapping of 150 breast cancer risk regions identifies 191 likely target genes. Nat. Genet., 52: 56–73

[203] Dzida,T., Iqbal,M., Charapitsa,I., Reid,G., Stunnenberg,H., Matarese,F., Grote,K., Honkela,A. (2017). Predicting stimulation-dependent enhancer-promoter interactions from ChIP-Seq time course data. PeerJ, 5: e3742

[204] Feng,Z. X., Li,Q. Z. Meng,J. (2018). Recognition of the long range enhancer-promoter interactions by further adding DNA structure properties and transcription factor binding motifs in human cell lines. J. Theor. Biol., 445: 136–150

[205] Hait,T. A., Elkon,R. (2022). CT-FOCS: a novel method for inferring cell type-specific enhancer-promoter maps. Nucleic Acids Res., 50: e55

RIGHTS & PERMISSIONS

The Author(s). Published by Higher Education Press.