LeNup: learning nucleosome positioning from DNA sequences with improved convolutional neural networks

The importance of DNA sequence for nucleosome positioning in transcriptional regulation

Nucleosome positioning is a key factor for transcriptional regulation. Nucleosomes regulate the dynamic accessibility of chromatin and interact with the transcription machinery at every stage. Influences to steer nucleosome positioning are diverse, and the according importance of the DNA sequence in contrast to active chromatin remodeling has been the subject of long discussion. In this study, we evaluate the functional role of DNA sequence for all major elements along the process of transcription. We developed a random forest classifier based on local DNA structure that assesses the sequence-intrinsic support for nucleosome positioning. On this basis, we created a simple data resource that we applied genome-wide to the human genome. In our comprehensive analysis, we found a special role of DNA in mediating the competition of nucleosomes with cis-regulatory elements, in enabling steady transcription, for positioning of stable nucleosomes in exons, and for repelling nucleosomes during transcription termination. In contrast, we relate these findings to concurrent processes that generate strongly positioned nucleosomes in vivo that are not mediated by sequence, such as energy-dependent remodeling of chromatin.

Interpretable deep residual network uncovers nucleosome positioning and associated features

Nucleosomes represent elementary building units of eukaryotic chromosomes and consist of DNA wrapped around a histone octamer flanked by linker DNA segments. Nucleosomes are central in epigenetic pathways and their genomic positioning is associated with regulation of gene expression, DNA replication, DNA methylation and DNA repair, among other functions. Building on prior discoveries that DNA sequences noticeably affect nucleosome positioning, our objective is to identify nucleosome positions and related features across entire genome. Here, we introduce an interpretable framework based on the concepts of deep residual networks (NuPoSe). Trained on high-coverage human experimental MNase-seq data, NuPoSe is able to learn sequence and structural patterns associated with nucleosome organization in human genome. NuPoSe can be also applied to unseen data from different organisms and cell types. Our findings point to 43 informative features, most of them constitute tri-nucleotides, di-nucleotides and one tetra-nucleotide. Most features are significantly associated with the nucleosomal structural characteristics, namely, periodicity of nucleosomal DNA and its location with respect to a histone octamer. Importantly, we show that features derived from the 27 bp linker DNA flanking nucleosomes contribute up to 10% to the quality of the prediction model. This, along with the comprehensive training sets, deep-learning architecture, and feature selection method, may contribute to the NuPoSe's 80-89% classification accuracy on different independent datasets.

CatNet: Sequence-based deep learning with cross-attention mechanism for identifying endocrine-disrupting chemicals

Endocrine-disrupting chemicals (EDCs) pose significant environmental and health risks due to their potential to interfere with nuclear receptors (NRs), key regulators of physiological processes. Despite the evident risks, the majority of existing research narrows its focus on the interaction between compounds and the individual NR target, neglecting a comprehensive assessment across the entire NR family. In response, this study assembled a comprehensive human NR dataset, capturing 49,244 interactions between 35,467 unique compounds and 42 NRs. We introduced a cross-attention network framework, "CatNet", innovatively integrating compound and protein representations through cross-attention mechanisms. The results showed that CatNet model achieved excellent performance with an area under the receiver operating characteristic curve (AUCROC) = 0.916 on the test set, and exhibited reliable generalization on unseen compound-NR pairs. A distinguishing feature of our research is its capacity to expand to novel targets. Beyond its predictive accuracy, CatNet offers a valuable mechanistic perspective on compound-NR interactions through feature visualization. Augmenting the utility of our research, we have also developed a graphical user interface, empowering researchers to predict chemical binding to diverse NRs. Our model enables the prediction of human NR-related EDCs and shows the potential to identify EDCs related to other targets.

Comprehensive computational analysis of epigenetic descriptors affecting CRISPR-Cas9 off-target activity

BACKGROUND

A common issue in CRISPR-Cas9 genome editing is off-target activity, which prevents the widespread use of CRISPR-Cas9 in medical applications. Among other factors, primary chromatin structure and epigenetics may influence off-target activity.

METHODS

In this work, we utilize crisprSQL, an off-target database, to analyze the effect of 19 epigenetic descriptors on CRISPR-Cas9 off-target activity. Termed as 19 epigenetic features/scores, they consist of 6 experimental epigenetic and 13 computed nucleosome organization-related features. In terms of novel features, 15 of the epigenetic scores are newly considered. The 15 newly considered scores consist of 13 freshly computed nucleosome occupancy/positioning scores and 2 experimental features (MNase and DRIP). The other 4 existing scores are experimental features (CTCF, DNase I, H3K4me3, RRBS) commonly used in deep learning models for off-target activity prediction. For data curation, MNase was aggregated from existing experimental nucleosome occupancy data. Based on the sequence context information available in crisprSQL, we also computed nucleosome occupancy/positioning scores for off-target sites.

RESULTS

To investigate the relationship between the 19 epigenetic features and off-target activity, we first conducted Spearman and Pearson correlation analysis. Such analysis shows that some computed scores derived from training-based models and training-free algorithms outperform all experimental epigenetic features. Next, we evaluated the contribution of all epigenetic features in two successful machine/deep learning models which predict off-target activity. We found that some computed scores, unlike all 6 experimental features, significantly contribute to the predictions of both models. As a practical research contribution, we make the off-target dataset containing all 19 epigenetic features available to the research community.

CONCLUSIONS

Our comprehensive computational analysis helps the CRISPR-Cas9 community better understand the relationship between epigenetic features and CRISPR-Cas9 off-target activity.

Assessment and Optimization of Explainable Machine Learning Models Applied to Transcriptomic Data

Explainable artificial intelligence aims to interpret how machine learning models make decisions, and many model explainers have been developed in the computer vision field. However, understanding of the applicability of these model explainers to biological data is still lacking. In this study, we comprehensively evaluated multiple explainers by interpreting pre-trained models for predicting tissue types from transcriptomic data and by identifying the top contributing genes from each sample with the greatest impacts on model prediction. To improve the reproducibility and interpretability of results generated by model explainers, we proposed a series of optimization strategies for each explainer on two different model architectures of multilayer perceptron (MLP) and convolutional neural network (CNN). We observed three groups of explainer and model architecture combinations with high reproducibility. Group II, which contains three model explainers on aggregated MLP models, identified top contributing genes in different tissues that exhibited tissue-specific manifestation and were potential cancer biomarkers. In summary, our work provides novel insights and guidance for exploring biological mechanisms using explainable machine learning models.

Genomics enters the deep learning era

The tremendous amount of biological sequence data available, combined with the recent methodological breakthrough in deep learning in domains such as computer vision or natural language processing, is leading today to the transformation of bioinformatics through the emergence of deep genomics, the application of deep learning to genomic sequences. We review here the new applications that the use of deep learning enables in the field, focusing on three aspects: the functional annotation of genomes, the sequence determinants of the genome functions and the possibility to write synthetic genomic sequences.

Galaxy Dnpatterntools for Computational Analysis of Nucleosome Positioning Sequence Patterns

Nucleosomes are basic units of DNA packing in eukaryotes. Their structure is well conserved from yeast to human and consists of the histone octamer core and 147 bp DNA wrapped around it. Nucleosomes are bound to a majority of the eukaryotic genomic DNA, including its regulatory regions. Hence, they also play a major role in gene regulation. For the latter, their precise positioning on DNA is essential. In the present paper, we describe Galaxy dnpatterntools—software package for nucleosome DNA sequence analysis and mapping. This software will be useful for computational biologists practitioners to conduct more profound studies of gene regulatory mechanisms.

Nucleosome positioning based on DNA sequence embedding and deep learning

Background

Nucleosome positioning is the precise determination of the location of nucleosomes on DNA sequence. With the continuous advancement of biotechnology and computer technology, biological data is showing explosive growth. It is of practical significance to develop an efficient nucleosome positioning algorithm. Indeed, convolutional neural networks (CNN) can capture local features in DNA sequences, but ignore the order of bases. While the bidirectional recurrent neural network can make up for CNN's shortcomings in this regard and extract the long-term dependent features of DNA sequence.

Results

In this work, we use word vectors to represent DNA sequences and propose three new deep learning models for nucleosome positioning, and the integrative model NP_CBiR reaches a better prediction performance. The overall accuracies of NP_CBiR on H. sapiens, C. elegans, and D. melanogaster datasets are 86.18%, 89.39%, and 85.55% respectively.

Conclusions

Benefited by different network structures, NP_CBiR can effectively extract local features and bases order features of DNA sequences, thus can be considered as a complementary tool for nucleosome positioning.

OUP accepted manuscript

DNA mechanical properties play a critical role in every aspect of DNA-dependent biological processes. Recently a high throughput assay named loop-seq has been developed to quantify the intrinsic bendability of a massive number of DNA fragments simultaneously. Using the loop-seq data, we develop a software tool, DNAcycP, based on a deep-learning approach for intrinsic DNA cyclizability prediction. We demonstrate DNAcycP predicts intrinsic DNA cyclizability with high fidelity compared to the experimental data. Using an independent dataset from in vitro selection for enrichment of loopable sequences, we further verified the predicted cyclizability score, termed C-score, can well distinguish DNA fragments with different loopability. We applied DNAcycP to multiple species and compared the C-scores with available high-resolution chemical nucleosome maps. Our analyses showed that both yeast and mouse genomes share a conserved feature of high DNA bendability spanning nucleosome dyads. Additionally, we extended our analysis to transcription factor binding sites and surprisingly found that the cyclizability is substantially elevated at CTCF binding sites in the mouse genome. We further demonstrate this distinct mechanical property is conserved across mammalian species and is inherent to CTCF binding DNA motif.

Identifying Molecular Biomarkers for Diseases With Machine Learning Based on Integrative Omics

Molecular biomarkers are certain molecules or set of molecules that can be of help for diagnosis or prognosis of diseases or disorders. In the past decades, thanks to the advances in high-throughput technologies, a huge amount of molecular 'omics' data, e.g., transcriptomics and proteomics, have been accumulated. The availability of these omics data makes it possible to screen biomarkers for diseases or disorders. Accordingly, a number of computational approaches have been developed to identify biomarkers by exploring the omics data. In this review, we present a comprehensive survey on the recent progress of identification of molecular biomarkers with machine learning approaches. Specifically, we categorize the machine learning approaches into supervised, un-supervised and recommendation approaches, where the biomarkers including single genes, gene sets and small gene networks. In addition, we further discuss potential problems underlying bio-medical data that may pose challenges for machine learning, and provide possible directions for future biomarker identification.

Identifying complex motifs in massive omics data with a variable-convolutional layer in deep neural network

Motif identification is among the most common and essential computational tasks for bioinformatics and genomics. Here we proposed a novel convolutional layer for deep neural network, named variable convolutional (vConv) layer, for effective motif identification in high-throughput omics data by learning kernel length from data adaptively. Empirical evaluations on DNA-protein binding and DNase footprinting cases well demonstrated that vConv-based networks have superior performance to their convolutional counterparts regardless of model complexity. Meanwhile, vConv could be readily integrated into multi-layer neural networks as an 'in-place replacement' of canonical convolutional layer. All source codes are freely available on GitHub for academic usage.

Comparative analysis and prediction of nucleosome positioning using integrative feature representation and machine learning algorithms

Background

Nucleosome plays an important role in the process of genome expression, DNA replication, DNA repair and transcription. Therefore, the research of nucleosome positioning has invariably received extensive attention. Considering the diversity of DNA sequence representation methods, we tried to integrate multiple features to analyze its effect in the process of nucleosome positioning analysis. This process can also deepen our understanding of the theoretical analysis of nucleosome positioning.

Results

Here, we not only used frequency chaos game representation (FCGR) to construct DNA sequence features, but also integrated it with other features and adopted the principal component analysis (PCA) algorithm. Simultaneously, support vector machine (SVM), extreme learning machine (ELM), extreme gradient boosting (XGBoost), multilayer perceptron (MLP) and convolutional neural networks (CNN) are used as predictors for nucleosome positioning prediction analysis, respectively. The integrated feature vector prediction quality is significantly superior to a single feature. After using principal component analysis (PCA) to reduce the feature dimension, the prediction quality of H. sapiens dataset has been significantly improved.

Conclusions

Comparative analysis and prediction on H. sapiens, C. elegans, D. melanogaster and S. cerevisiae datasets, demonstrate that the application of FCGR to nucleosome positioning is feasible, and we also found that integrative feature representation would be better.

Genome-wide prediction of DNA mutation effect on nucleosome positions for yeast synthetic genomics

Genetically modified genomes are often used today in many areas of fundamental and applied research. In many studies, coding or noncoding regions are modified in order to change protein sequences or gene expression levels. Modifying one or several nucleotides in a genome can also lead to unexpected changes in the epigenetic regulation of genes. When designing a synthetic genome with many mutations, it would thus be very informative to be able to predict the effect of these mutations on chromatin. We develop here a deep learning approach that quantifies the effect of every possible single mutation on nucleosome positions on the full Saccharomyces cerevisiae genome. This type of annotation track can be used when designing a modified S. cerevisiae genome. We further highlight how this track can provide new insights on the sequence-dependent mechanisms that drive nucleosomes' positions in vivo.

An Integrative Framework for Combining Sequence and Epigenomic Data to Predict Transcription Factor Binding Sites Using Deep Learning

Knowing the transcription factor binding sites (TFBSs) is essential for modeling the underlying binding mechanisms and follow-up cellular functions. Convolutional neural networks (CNNs) have outperformed methods in predicting TFBSs from the primary DNA sequence. In addition to DNA sequences, histone modifications and chromatin accessibility are also important factors influencing their activity. They have been explored to predict TFBSs recently. However, current methods rarely take into account histone modifications and chromatin accessibility using CNN in an integrative framework. To this end, we developed a general CNN model to integrate these data for predicting TFBSs. We systematically benchmarked a series of architecture variants by changing network structure in terms of width and depth, and explored the effects of sample length at flanking regions. We evaluated the performance of the three types of data and their combinations using 256 ChIP-seq experiments and also compared it with competing machine learning methods. We find that contributions from these three types of data are complementary to each other. Moreover, the integrative CNN framework is superior to traditional machine learning methods with significant improvements.

CORENup: a combination of convolutional and recurrent deep neural networks for nucleosome positioning identification

BACKGROUND

Nucleosomes wrap the DNA into the nucleus of the Eukaryote cell and regulate its transcription phase. Several studies indicate that nucleosomes are determined by the combined effects of several factors, including DNA sequence organization. Interestingly, the identification of nucleosomes on a genomic scale has been successfully performed by computational methods using DNA sequence as input data.

RESULTS

In this work, we propose CORENup, a deep learning model for nucleosome identification. CORENup processes a DNA sequence as input using one-hot representation and combines in a parallel fashion a fully convolutional neural network and a recurrent layer. These two parallel levels are devoted to catching both non periodic and periodic DNA string features. A dense layer is devoted to their combination to give a final classification.

CONCLUSIONS

Results computed on public data sets of different organisms show that CORENup is a state of the art methodology for nucleosome positioning identification based on a Deep Neural Network architecture. The comparisons have been carried out using two groups of datasets, currently adopted by the best performing methods, and CORENup has shown top performance both in terms of classification metrics and elapsed computation time.

BITS2019: the sixteenth annual meeting of the Italian society of bioinformatics

The 16th Annual Meeting of the Bioinformatics Italian Society was held in Palermo, Italy, on June 26-28, 2019. More than 80 scientific contributions were presented, including 4 keynote lectures, 31 oral communications and 49 posters. Also, three workshops were organised before and during the meeting. Full papers from some of the works presented in Palermo were submitted for this Supplement of BMC Bioinformatics. Here, we provide an overview of meeting aims and scope. We also shortly introduce selected papers that have been accepted for publication in this Supplement, for a complete presentation of the outcomes of the meeting.

[Identification of nucleosome positioning using support vector machine method based on comprehensive DNA sequence feature]

In this article, based on z-curve theory and position weight matrix (PWM), a model for nucleosome sequences was constructed. Nucleosome sequence dataset was transformed into three-dimensional coordinates, PWM of the nucleosome sequences was calculated and the similarity score was obtained. After integrating them, a nucleosome feature model based on the comprehensive DNA sequences was obtained and named CSeqFM. We calculated the Euclidean distance between nucleosome sequence candidates or linker sequences and CSeqFM model as the feature dataset, and put the feature datasets into the support vector machine (SVM) for training and testing by ten-fold cross-validation. The results showed that the sensitivity, specificity, accuracy and Matthews correlation coefficient (MCC) of identifying nucleosome positioning for S. cerevisiae were 97.1%, 96.9%, 94.2% and 0.89, respectively, and the area under the receiver operating characteristic curve (AUC) was 0.980 1. Compared with another z-curve method, it was found that our method had better identifying effect and each evaluation performance showed better superiority. CSeqFM method was applied to identify nucleosome positioning for other three species, including C. elegans, H. sapiens and D. melanogaster. The results showed that AUCs of the three species were all higher than 0.90, and CSeqFM method also showed better stability and effectiveness compared with iNuc-STNC and iNuc-PseKNC methods, which is further demonstrated that CSeqFM method has strong reliability and good identification performance.

NucMap: a database of genome-wide nucleosome positioning map across species

Dynamics of nucleosome positioning affects chromatin state, transcription and all other biological processes occurring on genomic DNA. While MNase-Seq has been used to depict nucleosome positioning map in eukaryote in the past years, nucleosome positioning data is increasing dramatically. To facilitate the usage of published data across studies, we developed a database named nucleosome positioning map (NucMap, http://bigd.big.ac.cn/nucmap). NucMap includes 798 experimental data from 477 samples across 15 species. With a series of functional modules, users can search profile of nucleosome positioning at the promoter region of each gene across all samples and make enrichment analysis on nucleosome positioning data in all genomic regions. Nucleosome browser was built to visualize the profiles of nucleosome positioning. Users can also visualize multiple sources of omics data with the nucleosome browser and make side-by-side comparisons. All processed data in the database are freely available. NucMap is the first comprehensive nucleosome positioning platform and it will serve as an important resource to facilitate the understanding of chromatin regulation.

Prediction of sgRNA on-target activity in bacteria by deep learning

BACKGROUND

One of the main challenges for the CRISPR-Cas9 system is selecting optimal single-guide RNAs (sgRNAs). Recently, deep learning has enhanced sgRNA prediction in eukaryotes. However, the prokaryotic chromatin structure is different from eukaryotes, so models trained on eukaryotes may not apply to prokaryotes.

RESULTS

We designed and implemented a convolutional neural network to predict sgRNA activity in Escherichia coli. The network was trained and tested on the recently-released sgRNA activity dataset. Our convolutional neural network achieved excellent performance, yielding average Spearman correlation coefficients of 0.5817, 0.7105, and 0.3602, respectively for Cas9, eSpCas9 and Cas9 with a recA coding region deletion. We confirmed that the sgRNA prediction models trained on prokaryotes do not apply to eukaryotes and vice versa. We adopted perturbation-based approaches to analyze distinct biological patterns between prokaryotic and eukaryotic editing. Then, we improved the predictive performance of the prokaryotic Cas9 system by transfer learning. Finally, we determined that potential off-target scores accumulated on a genome-wide scale affect on-target activity, which could slightly improve on-target predictive performance.

CONCLUSIONS

We developed convolutional neural networks to predict sgRNA activity for wild type and mutant Cas9 in prokaryotes. Our results show that the prediction accuracy of our method is improved over state-of-the-art models.

ZCMM: A Novel Method Using Z-Curve Theory- Based and Position Weight Matrix for Predicting Nucleosome Positioning

Nucleosomes are the basic units of eukaryotes. The accurate positioning of nucleosomes plays a significant role in understanding many biological processes such as transcriptional regulation mechanisms and DNA replication and repair. Here, we describe the development of a novel method, termed ZCMM, based on Z-curve theory and position weight matrix (PWM). The ZCMM was trained and tested using the nucleosomal and linker sequences determined by support vector machine (SVM) in Saccharomyces cerevisiae (S. cerevisiae), and experimental results showed that the sensitivity (Sn), specificity (Sp), accuracy (Acc), and Matthews correlation coefficient (MCC) values for ZCMM were 91.40%, 96.56%, 96.75%, and 0.88, respectively, and the average area under the receiver operating characteristic curve (AUC) value was 0.972. A ZCMM predictor was developed to predict nucleosome positioning in Homo sapiens (H. sapiens), Caenorhabditis elegans (C. elegans), and Drosophila melanogaster (D. melanogaster) genomes, and the accuracy (Acc) values were 77.72%, 85.34%, and 93.62%, respectively. The maximum AUC values of the four species were 0.982, 0.861, 0.912 and 0.911, respectively. Another independent dataset for S. cerevisiae was used to predict nucleosome positioning. Compared with the results of Wu's method, it was found that the Sn, Sp, Acc, and MCC of ZCMM results for S. cerevisiae were all higher, reaching 96.72%, 96.54%, 94.10%, and 0.88. Compared with the Guo's method 'iNuc-PseKNC', the results of ZCMM for D. melanogaster were better. Meanwhile, the ZCMM was compared with some experimental data in vitro and in vivo for S. cerevisiae, and the results showed that the nucleosomes predicted by ZCMM were highly consistent with those confirmed by these experiments. Therefore, it was further confirmed that the ZCMM method has good accuracy and reliability in predicting nucleosome positioning.

Deep learning for DNase I hypersensitive sites identification

Background

The DNase I hypersensitive sites (DHSs) are associated with the cis-regulatory DNA elements. An efficient method of identifying DHSs can enhance the understanding on the accessibility of chromatin. Despite a multitude of resources available on line including experimental datasets and computational tools, the complex language of DHSs remains incompletely understood.

Methods

Here, we address this challenge using an approach based on a state-of-the-art machine learning method. We present a novel convolutional neural network (CNN) which combined Inception like networks with a gating mechanism for the response of multiple patterns and longterm association in DNA sequences to predict multi-scale DHSs in Arabidopsis, rice and Homo sapiens.

Results

Our method obtains 0.961 area under curve (AUC) on Arabidopsis, 0.969 AUC on rice and 0.918 AUC on Homo sapiens.

Conclusions

Our method provides an efficient and accurate way to identify multi-scale DHSs sequences by deep learning.

Rise of Deep Learning for Genomic, Proteomic, and Metabolomic Data Integration in Precision Medicine

Machine learning (ML) is being ubiquitously incorporated into everyday products such as Internet search, email spam filters, product recommendations, image classification, and speech recognition. New approaches for highly integrated manufacturing and automation such as the Industry 4.0 and the Internet of things are also converging with ML methodologies. Many approaches incorporate complex artificial neural network architectures and are collectively referred to as deep learning (DL) applications. These methods have been shown capable of representing and learning predictable relationships in many diverse forms of data and hold promise for transforming the future of omics research and applications in precision medicine. Omics and electronic health record data pose considerable challenges for DL. This is due to many factors such as low signal to noise, analytical variance, and complex data integration requirements. However, DL models have already been shown capable of both improving the ease of data encoding and predictive model performance over alternative approaches. It may not be surprising that concepts encountered in DL share similarities with those observed in biological message relay systems such as gene, protein, and metabolite networks. This expert review examines the challenges and opportunities for DL at a systems and biological scale for a precision medicine readership.