Comprehensive evaluation of deep learning architectures for prediction of DNA/RNA sequence binding specificities

Big data and deep learning for RNA biology

The exponential growth of big data in RNA biology (RB) has led to the development of deep learning (DL) models that have driven crucial discoveries. As constantly evidenced by DL studies in other fields, the successful implementation of DL in RB depends heavily on the effective utilization of large-scale datasets from public databases. In achieving this goal, data encoding methods, learning algorithms, and techniques that align well with biological domain knowledge have played pivotal roles. In this review, we provide guiding principles for applying these DL concepts to various problems in RB by demonstrating successful examples and associated methodologies. We also discuss the remaining challenges in developing DL models for RB and suggest strategies to overcome these challenges. Overall, this review aims to illuminate the compelling potential of DL for RB and ways to apply this powerful technology to investigate the intriguing biology of RNA more effectively.

Improving plant miRNA-target prediction with self-supervised k-mer embedding and spectral graph convolutional neural network

Deciphering the targets of microRNAs (miRNAs) in plants is crucial for comprehending their function and the variation in phenotype that they cause. As the highly cell-specific nature of miRNA regulation, recent computational approaches usually utilize expression data to identify the most physiologically relevant targets. Although these methods are effective, they typically require a large sample size and high-depth sequencing to detect potential miRNA-target pairs, thereby limiting their applicability in improving plant breeding. In this study, we propose a novel miRNA-target prediction framework named kmerPMTF (k-mer-based prediction framework for plant miRNA-target). Our framework effectively extracts the latent semantic embeddings of sequences by utilizing k-mer splitting and a deep self-supervised neural network. We construct multiple similarity networks based on k-mer embeddings and employ graph convolutional networks to derive deep representations of miRNAs and targets and calculate the probabilities of potential associations. We evaluated the performance of kmerPMTF on four typical plant datasets: Arabidopsis thaliana, Oryza sativa, Solanum lycopersicum, and Prunus persica. The results demonstrate its ability to achieve AUPRC values of 84.9%, 91.0%, 80.1%, and 82.1% in 5-fold cross-validation, respectively. Compared with several state-of-the-art existing methods, our framework achieves better performance on threshold-independent evaluation metrics. Overall, our study provides an efficient and simplified methodology for identifying plant miRNA-target associations, which will contribute to a deeper comprehension of miRNA regulatory mechanisms in plants.

Deep Learning for Elucidating Modifications to RNA-Status and Challenges Ahead

RNA-binding proteins and chemical modifications to RNA play vital roles in the co- and post-transcriptional regulation of genes. In order to fully decipher their biological roles, it is an essential task to catalogue their precise target locations along with their preferred contexts and sequence-based determinants. Recently, deep learning approaches have significantly advanced in this field. These methods can predict the presence or absence of modification at specific genomic regions based on diverse features, particularly sequence and secondary structure, allowing us to decipher the highly non-linear sequence patterns and structures that underlie site preferences. This article provides an overview of how deep learning is being applied to this area, with a particular focus on the problem of mRNA-RBP binding, while also considering other types of chemical modification to RNA. It discusses how different types of model can handle sequence-based and/or secondary-structure-based inputs, the process of model training, including choice of negative regions and separating sets for testing and training, and offers recommendations for developing biologically relevant models. Finally, it highlights four key areas that are crucial for advancing the field.

Prioritization of genes for translation: a computational approach

INTRODUCTION

Gene identification for genetic diseases is critical for the development of new diagnostic approaches and personalized treatment options. Prioritization of gene translation is an important consideration in the molecular biology field, allowing researchers to focus on the most promising candidates for further investigation.

AREAS COVERED

In this paper, we discussed different approaches to prioritize genes for translation, including the use of computational tools and machine learning algorithms, as well as experimental techniques such as knockdown and overexpression studies. We also explored the potential biases and limitations of these approaches and proposed strategies to improve the accuracy and reliability of gene prioritization methods. Although numerous computational methods have been developed for this purpose, there is a need for computational methods that incorporate tissue-specific information to enable more accurate prioritization of candidate genes. Such methods should provide tissue-specific predictions, insights into underlying disease mechanisms, and more accurate prioritization of genes.

EXPERT OPINION

Using advanced computational tools and machine learning algorithms to prioritize genes, we can identify potential targets for therapeutic intervention of complex diseases. This represents an up-and-coming method for drug development and personalized medicine.

Seq-RBPPred: Predicting RNA-Binding Proteins from Sequence

RNA-binding proteins (RBPs) can interact with RNAs to regulate RNA translation, modification, splicing, and other important biological processes. The accurate identification of RBPs is of paramount importance for gaining insights into the intricate mechanisms underlying organismal life activities. Traditional experimental methods to predict RBPs require a lot of time and money, so it is important to develop computational methods to predict RBPs. However, the existing approaches for RBP prediction still require further improvement due to unidentified RBPs in many species. In this study, we present Seq-RBPPred (predicting RBPs from sequence), a novel method that utilizes a comprehensive feature representation encompassing both biophysical properties and hidden-state features derived from protein sequences. In the results, comprehensive performance evaluations of Seq-RBPPred its superiority compare with state-of-the-art methods, yielding impressive performance including 0.922 for overall accuracy, 0.926 for sensitivity, 0.903 for specificity, and Matthew's correlation coefficient (MCC) of 0.757 as ascertained from the evaluation of the testing set. The data and code of Seq-RBPPred are available at https://github.com/yaoyao-11/Seq-RBPPred.

Decoding protein binding landscape on circular RNAs with base-resolution transformer models

Circular RNAs (circRNAs), a class of endogenous RNA with a covalent loop structure, can regulate gene expression by serving as sponges for microRNAs and RNA-binding proteins (RBPs). To date, most computational methods for predicting RBP binding sites on circRNAs focus on circRNA fragments instead of circRNAs. These methods detect whether a circRNA fragment contains binding sites, but cannot determine where are the binding sites and how many binding sites are on the circRNA transcript. We report a hybrid deep learning-based tool, CircSite, to predict RBP binding sites at single-nucleotide resolution and detect key contributed nucleotides on circRNA transcripts. CircSite takes advantage of convolutional neural networks (CNNs) and Transformer for learning local and global representations of circRNAs binding to RBPs, respectively. We construct 37 datasets of circRNAs interacting with proteins for benchmarking and the experimental results show that CircSite offers accurate predictions of RBP binding nucleotides and detects key subsequences aligning well with known binding motifs. CircSite is an easy-to-use online webserver for predicting RBP binding sites on circRNA transcripts and freely available at http://www.csbio.sjtu.edu.cn/bioinf/CircSite/.

Predicting circRNA-miRNA interactions utilizing transformer-based RNA sequential learning and high-order proximity preserved embedding

A key regulatory mechanism involves circular RNA (circRNA) acting as a sponge to modulate microRNA (miRNA), and thus, studying their interaction has significant medical implications. In this field, there are currently two pressing issues that remain unresolved. Firstly, due to the scarcity of verified interactions, we require a minimal amount of samples for training. Secondly, the current models lack interpretability. Therefore, we propose SPBCMI, a method that combines sequence features extracted using the Bidirectional Encoder Representations from Transformer (BERT) model and structural features of biological molecule networks extracted through graph embedding to train a GBDT (Gradient-boosted decision trees) classifier for prediction. Our method yielded an AUC of 0.9143, which is currently the best for this problem. Furthermore, in the case study, SPBCMI accurately predicted 7 out of 10 circRNA-miRNA interactions. These results show that our method provides an innovative and high-performing approach to understanding the interaction between circRNA and miRNA.

Discovery of a non-canonical GRHL1 binding site using deep convolutional and recurrent neural networks

BACKGROUND

Transcription factors regulate gene expression by binding to transcription factor binding sites (TFBSs). Most models for predicting TFBSs are based on position weight matrices (PWMs), which require a specific motif to be present in the DNA sequence and do not consider interdependencies of nucleotides. Novel approaches such as Transcription Factor Flexible Models or recurrent neural networks consequently provide higher accuracies. However, it is unclear whether such approaches can uncover novel non-canonical, hitherto unexpected TFBSs relevant to human transcriptional regulation.

RESULTS

In this study, we trained a convolutional recurrent neural network with HT-SELEX data for GRHL1 binding and applied it to a set of GRHL1 binding sites obtained from ChIP-Seq experiments from human cells. We identified 46 non-canonical GRHL1 binding sites, which were not found by a conventional PWM approach. Unexpectedly, some of the newly predicted binding sequences lacked the CNNG core motif, so far considered obligatory for GRHL1 binding. Using isothermal titration calorimetry, we experimentally confirmed binding between the GRHL1-DNA binding domain and predicted GRHL1 binding sites, including a non-canonical GRHL1 binding site. Mutagenesis of individual nucleotides revealed a correlation between predicted binding strength and experimentally validated binding affinity across representative sequences. This correlation was neither observed with a PWM-based nor another deep learning approach.

CONCLUSIONS

Our results show that convolutional recurrent neural networks may uncover unanticipated binding sites and facilitate quantitative transcription factor binding predictions.

Interpretable prediction models for widespread m6A RNA modification across cell lines and tissues

MOTIVATION

RNA N6-methyladenosine (m6A) in Homo sapiens plays vital roles in a variety of biological functions. Precise identification of m6A modifications is thus essential to elucidation of their biological functions and underlying molecular-level mechanisms. Currently available high-throughput single-nucleotide-resolution m6A modification data considerably accelerated the identification of RNA modification sites through the development of data-driven computational methods. Nevertheless, existing methods have limitations in terms of the coverage of single-nucleotide-resolution cell lines and have poor capability in model interpretations, thereby having limited applicability.

RESULTS

In this study, we present CLSM6A, comprising a set of deep learning-based models designed for predicting single-nucleotide-resolution m6A RNA modification sites across eight different cell lines and three tissues. Extensive benchmarking experiments are conducted on well-curated datasets and accordingly, CLSM6A achieves superior performance than current state-of-the-art methods. Furthermore, CLSM6A is capable of interpreting the prediction decision-making process by excavating critical motifs activated by filters and pinpointing highly concerned positions in both forward and backward propagations. CLSM6A exhibits better portability on similar cross-cell line/tissue datasets, reveals a strong association between highly activated motifs and high-impact motifs, and demonstrates complementary attributes of different interpretation strategies.

AVAILABILITY AND IMPLEMENTATION

The webserver is available at http://csbio.njust.edu.cn/bioinf/clsm6a. The datasets and code are available at https://github.com/zhangying-njust/CLSM6A/.

Artificial intelligence: Machine learning approach for screening large database and drug discovery

Recent research in drug discovery dealing with many faces difficulties, including development of new drugs during disease outbreak and drug resistance due to rapidly accumulating mutations. Virtual screening is the most widely used method in computer aided drug discovery. It has a prominent ability in screening drug targets from large molecular databases. Recently, a number of web servers have developed for quickly screening publicly accessible chemical databases. In a nutshell, deep learning algorithms and artificial neural networks have modernised the field. Several drug discovery processes have used machine learning and deep learning algorithms, including peptide synthesis, structure-based virtual screening, ligand-based virtual screening, toxicity prediction, drug monitoring and release, pharmacophore modelling, quantitative structure-activity relationship, drug repositioning, polypharmacology, and physiochemical activity. Although there are presently a wide variety of data-driven AI/ML tools available, the majority of these tools have, up to this point, been developed in the context of non-communicable diseases like cancer, and a number of obstacles have prevented the translation of these tools to the discovery of treatments against infectious diseases. In this review various aspects of AI and ML in virtual screening of large databases were discussed. Here, with an emphasis on antivirals as well as other disease, offers a perspective on the advantages, drawbacks, and hazards of AI/ML techniques in the search for innovative treatments.

KDeep: a new memory-efficient data extraction method for accurately predicting DNA/RNA transcription factor binding sites

This paper addresses the crucial task of identifying DNA/RNA binding sites, which has implications in drug/vaccine design, protein engineering, and cancer research. Existing methods utilize complex neural network structures, diverse input types, and machine learning techniques for feature extraction. However, the growing volume of sequences poses processing challenges. This study introduces KDeep, employing a CNN-LSTM architecture with a novel encoding method called 2Lk. 2Lk enhances prediction accuracy, reduces memory consumption by up to 84%, reduces trainable parameters, and improves interpretability by approximately 79% compared to state-of-the-art approaches. KDeep offers a promising solution for accurate and efficient binding site prediction.

Benchmarking deep learning methods for predicting CRISPR/Cas9 sgRNA on- and off-target activities

In silico design of single guide RNA (sgRNA) plays a critical role in clustered regularly interspaced, short palindromic repeats/CRISPR-associated protein 9 (CRISPR/Cas9) system. Continuous efforts are aimed at improving sgRNA design with efficient on-target activity and reduced off-target mutations. In the last 5 years, an increasing number of deep learning-based methods have achieved breakthrough performance in predicting sgRNA on- and off-target activities. Nevertheless, it is worthwhile to systematically evaluate these methods for their predictive abilities. In this review, we conducted a systematic survey on the progress in prediction of on- and off-target editing. We investigated the performances of 10 mainstream deep learning-based on-target predictors using nine public datasets with different sample sizes. We found that in most scenarios, these methods showed superior predictive power on large- and medium-scale datasets than on small-scale datasets. In addition, we performed unbiased experiments to provide in-depth comparison of eight representative approaches for off-target prediction on 12 publicly available datasets with various imbalanced ratios of positive/negative samples. Most methods showed excellent performance on balanced datasets but have much room for improvement on moderate- and severe-imbalanced datasets. This study provides comprehensive perspectives on CRISPR/Cas9 sgRNA on- and off-target activity prediction and improvement for method development.

A systematic benchmark of machine learning methods for protein-RNA interaction prediction

RNA-binding proteins (RBPs) are central actors of RNA post-transcriptional regulation. Experiments to profile-binding sites of RBPs in vivo are limited to transcripts expressed in the experimental cell type, creating the need for computational methods to infer missing binding information. While numerous machine-learning based methods have been developed for this task, their use of heterogeneous training and evaluation datasets across different sets of RBPs and CLIP-seq protocols makes a direct comparison of their performance difficult. Here, we compile a set of 37 machine learning (primarily deep learning) methods for in vivo RBP-RNA interaction prediction and systematically benchmark a subset of 11 representative methods across hundreds of CLIP-seq datasets and RBPs. Using homogenized sample pre-processing and two negative-class sample generation strategies, we evaluate methods in terms of predictive performance and assess the impact of neural network architectures and input modalities on model performance. We believe that this study will not only enable researchers to choose the optimal prediction method for their tasks at hand, but also aid method developers in developing novel, high-performing methods by introducing a standardized framework for their evaluation.

WVDL: Weighted Voting Deep Learning Model for Predicting RNA-Protein Binding Sites

RNA-binding proteins are important for the process of cell life activities. High-throughput technique experimental method to discover RNA-protein binding sites is time-consuming and expensive. Deep learning is an effective theory for predicting RNA-protein binding sites. Using weighted voting method to integrate multiple basic classifier models can improve model performance. Thus, in our study, we propose a weighted voting deep learning model (WVDL), which uses weighted voting method to combine convolutional neural network (CNN), long short term memory network (LSTM) and residual network (ResNet). First, the final forecast result of WVDL outperforms the basic classifier models and other ensemble strategies. Second, WVDL can extract more effective features by using weighted voting to find the best weighted combination. And, the CNN model also can draw the predicted motif pictures. Third, WVDL gets a competitive experiment result on public RBP-24 datasets comparing with other state-of-the-art methods. The source code of our proposed WVDL can be found in https://github.com/biomg/WVDL.

A Bayesian approach to estimate MHC-peptide binding threshold

Major histocompatibility complex (MHC)-peptide binding is a critical step in enabling a peptide to serve as an antigen for T-cell recognition. Accurate prediction of this binding can facilitate various applications in immunotherapy. While many existing methods offer good predictive power for the binding affinity of a peptide to a specific MHC, few models attempt to infer the binding threshold that distinguishes binding sequences. These models often rely on experience-based ad hoc criteria, such as 500 or 1000nM. However, different MHCs may have different binding thresholds. As such, there is a need for an automatic, data-driven method to determine an accurate binding threshold. In this study, we proposed a Bayesian model that jointly infers core locations (binding sites), the binding affinity and the binding threshold. Our model provided the posterior distribution of the binding threshold, enabling accurate determination of an appropriate threshold for each MHC. To evaluate the performance of our method under different scenarios, we conducted simulation studies with varying dominant levels of motif distributions and proportions of random sequences. These simulation studies showed desirable estimation accuracy and robustness of our model. Additionally, when applied to real data, our results outperformed commonly used thresholds.

Characterization of epitranscriptome reader proteins experimentally and in silico: Current knowledge and future perspectives beyond the YTH domain

To date, over 150 chemical modifications to the four canonical RNA bases have been discovered, known collectively as the epitranscriptome. Many of these modifications have been implicated in a variety of cellular processes and disease states. Additional work has been done to identify proteins known as "readers" that selectively interact with RNAs that contain specific chemical modifications. Protein interactomes with N6-methyladenosine (m⁶A), N1-methyladenosine (m¹A), N5-methylcytosine (m⁵C), and 8-oxo-7,8-dihydroguanosine (8-oxoG) have been determined, mainly through experimental advances in proteomics techniques. However, relatively few proteins have been confirmed to bind directly to RNA containing these modifications. Furthermore, for many of these protein readers, the exact binding mechanisms as well as the exclusivity for recognition of modified RNA species remain elusive, leading to questions regarding their roles within different cellular processes. In the case of the YT-521B homology (YTH) family of proteins, both experimental and in silico techniques have been leveraged to provide valuable biophysical insights into the mechanisms of m⁶A recognition at atomic resolution. To date, the YTH family is one of the best characterized classes of readers. Here, we review current knowledge about epitranscriptome recognition of the YTH domain proteins from previously published experimental and computational studies. We additionally outline knowledge gaps for proteins beyond the well-studied human YTH domains and the current in silico techniques and resources that can enable investigation of protein interactions with modified RNA outside of the YTH-m⁶A context.

DeepAlgPro: an interpretable deep neural network model for predicting allergenic proteins

Allergies have become an emerging public health problem worldwide. The most effective way to prevent allergies is to find the causative allergen at the source and avoid re-exposure. However, most of the current computational methods used to identify allergens were based on homology or conventional machine learning methods, which were inefficient and still had room to be improved for the detection of allergens with low homology. In addition, few methods based on deep learning were reported, although deep learning has been successfully applied to several tasks in protein sequence analysis. In the present work, a deep neural network-based model, called DeepAlgPro, was proposed to identify allergens. We showed its great accuracy and applicability to large-scale forecasts by comparing it to other available tools. Additionally, we used ablation experiments to demonstrate the critical importance of the convolutional module in our model. Moreover, further analyses showed that epitope features contributed to model decision-making, thus improving the model's interpretability. Finally, we found that DeepAlgPro was capable of detecting potential new allergens. Overall, DeepAlgPro can serve as powerful software for identifying allergens.

CRBP-HFEF: Prediction of RBP-Binding Sites on circRNAs Based on Hierarchical Feature Expansion and Fusion

Circular RNAs (circRNAs) participate in the regulation of biological processes by binding to specific proteins and thus influence transcriptional processes. In recent years, circRNAs have become an emerging hotspot in RNA research. Due to powerful learning ability, the various deep learning frameworks have been used to predict the binding sites of RNA-binding protein (RPB) on circRNAs. These methods usually perform only single-level feature extraction of sequence information. However, the feature acquisition may be inadequate for single-level extraction. Generally, the features of deep and shallow layers of neural network can complement each other and are both important for binding site prediction tasks. Based on this concept, we propose a method that combines deep and shallow features, namely CRBP-HFEF. Specifically, features are first extracted and expanded for different levels of network. Then, the expanded deep and shallow features are fused and fed into the classification network, which finally determines whether they are binding sites. Compared to several existing methods, the experimental results on multiple datasets show that the proposed method achieves significant improvements in a number of metrics (with an average AUC of 0.9855). Moreover, much sufficient ablation experiments are also performed to verify the effectiveness of the hierarchical feature expansion strategy.

A survey on algorithms to characterize transcription factor binding sites

Transcription factors (TFs) are key regulatory proteins that control the transcriptional rate of cells by binding short DNA sequences called transcription factor binding sites (TFBS) or motifs. Identifying and characterizing TFBS is fundamental to understanding the regulatory mechanisms governing the transcriptional state of cells. During the last decades, several experimental methods have been developed to recover DNA sequences containing TFBS. In parallel, computational methods have been proposed to discover and identify TFBS motifs based on these DNA sequences. This is one of the most widely investigated problems in bioinformatics and is referred to as the motif discovery problem. In this manuscript, we review classical and novel experimental and computational methods developed to discover and characterize TFBS motifs in DNA sequences, highlighting their advantages and drawbacks. We also discuss open challenges and future perspectives that could fill the remaining gaps in the field.

Quantitative model for genome-wide cyclic AMP receptor protein binding site identification and characteristic analysis

Cyclic AMP receptor proteins (CRPs) are important transcription regulators in many species. The prediction of CRP-binding sites was mainly based on position-weighted matrixes (PWMs). Traditional prediction methods only considered known binding motifs, and their ability to discover inflexible binding patterns was limited. Thus, a novel CRP-binding site prediction model called CRPBSFinder was developed in this research, which combined the hidden Markov model, knowledge-based PWMs and structure-based binding affinity matrixes. We trained this model using validated CRP-binding data from Escherichia coli and evaluated it with computational and experimental methods. The result shows that the model not only can provide higher prediction performance than a classic method but also quantitatively indicates the binding affinity of transcription factor binding sites by prediction scores. The prediction result included not only the most knowns regulated genes but also 1089 novel CRP-regulated genes. The major regulatory roles of CRPs were divided into four classes: carbohydrate metabolism, organic acid metabolism, nitrogen compound metabolism and cellular transport. Several novel functions were also discovered, including heterocycle metabolic and response to stimulus. Based on the functional similarity of homologous CRPs, we applied the model to 35 other species. The prediction tool and the prediction results are online and are available at: https://awi.cuhk.edu.cn/∼CRPBSFinder.

Evidence for the role of transcription factors in the co-transcriptional regulation of intron retention

BACKGROUND

Alternative splicing is a widespread regulatory phenomenon that enables a single gene to produce multiple transcripts. Among the different types of alternative splicing, intron retention is one of the least explored despite its high prevalence in both plants and animals. The recent discovery that the majority of splicing is co-transcriptional has led to the finding that chromatin state affects alternative splicing. Therefore, it is plausible that transcription factors can regulate splicing outcomes.

RESULTS

We provide evidence for the hypothesis that transcription factors are involved in the regulation of intron retention by studying regions of open chromatin in retained and excised introns. Using deep learning models designed to distinguish between regions of open chromatin in retained introns and non-retained introns, we identified motifs enriched in IR events with significant hits to known human transcription factors. Our model predicts that the majority of transcription factors that affect intron retention come from the zinc finger family. We demonstrate the validity of these predictions using ChIP-seq data for multiple zinc finger transcription factors and find strong over-representation for their peaks in intron retention events.

CONCLUSIONS

This work opens up opportunities for further studies that elucidate the mechanisms by which transcription factors affect intron retention and other forms of splicing.

AVAILABILITY

Source code available at https://github.com/fahadahaf/chromir.

Identifying Disease of Interest With Deep Learning Using Diagnosis Code

BACKGROUND

Autoencoder (AE) is one of the deep learning techniques that uses an artificial neural network to reconstruct its input data in the output layer. We constructed a novel supervised AE model and tested its performance in the prediction of a co-existence of the disease of interest only using diagnostic codes.

METHODS

Diagnostic codes of one million randomly sampled patients listed in the Korean National Health Information Database in 2019 were used to train, validate, and test the prediction model. The first used AE solely for a feature engineering tool for an input of a classifier. Supervised Multi-Layer Perceptron (sMLP) was added to train a classifier to predict a binary level with latent representation as an input (AE + sMLP). The second model simultaneously updated the parameters in the AE and the connected MLP classifier during the learning process (End-to-End Supervised AE [EEsAE]). We tested the performances of these two models against baseline models, eXtreme Gradient Boosting (XGB) and naïve Bayes, in the prediction of co-existing gastric cancer diagnosis.

RESULTS

The proposed EEsAE model yielded the highest F1-score and highest area under the curve (0.86). The EEsAE and AE + sMLP gave the highest recalls. XGB yielded the highest precision. Ablation study revealed that iron deficiency anemia, gastroesophageal reflux disease, essential hypertension, gastric ulcers, benign prostate hyperplasia, and shoulder lesion were the top 6 most influential diagnoses on performance.

CONCLUSION

A novel EEsAE model showed promising performance in the prediction of a disease of interest.

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

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

RNAdegformer: accurate prediction of mRNA degradation at nucleotide resolution with deep learning

Messenger RNA-based therapeutics have shown tremendous potential, as demonstrated by the rapid development of messenger RNA based vaccines for COVID-19. Nevertheless, distribution of mRNA vaccines worldwide has been hampered by mRNA's inherent thermal instability due to in-line hydrolysis, a chemical degradation reaction. Therefore, predicting and understanding RNA degradation is a crucial and urgent task. Here we present RNAdegformer, an effective and interpretable model architecture that excels in predicting RNA degradation. RNAdegformer processes RNA sequences with self-attention and convolutions, two deep learning techniques that have proved dominant in the fields of computer vision and natural language processing, while utilizing biophysical features of RNA. We demonstrate that RNAdegformer outperforms previous best methods at predicting degradation properties at nucleotide resolution for COVID-19 mRNA vaccines. RNAdegformer predictions also exhibit improved correlation with RNA in vitro half-life compared with previous best methods. Additionally, we showcase how direct visualization of self-attention maps assists informed decision-making. Further, our model reveals important features in determining mRNA degradation rates via leave-one-feature-out analysis.

High-throughput biochemistry in RNA sequence space: predicting structure and function

RNAs are central to fundamental biological processes in all known organisms. The set of possible intramolecular interactions of RNA nucleotides defines the range of alternative structural conformations of a specific RNA that can coexist, and these structures enable functional catalytic properties of RNAs and/or their productive intermolecular interactions with other RNAs or proteins. However, the immense combinatorial space of potential RNA sequences has precluded predictive mapping between RNA sequence and molecular structure and function. Recent advances in high-throughput approaches in vitro have enabled quantitative thermodynamic and kinetic measurements of RNA-RNA and RNA-protein interactions, across hundreds of thousands of sequence variations. In this Review, we explore these techniques, how they can be used to understand RNA function and how they might form the foundations of an accurate model to predict the structure and function of an RNA directly from its nucleotide sequence. The experimental techniques and modelling frameworks discussed here are also highly relevant for the sampling of sequence-structure-function space of DNAs and proteins.

Predicting the Sequence Specificities of DNA-Binding Proteins by DNA Fine-Tuned Language Model With Decaying Learning Rates

DNA-binding proteins (DBPs) play vital roles in the regulation of biological systems. Although there are already many deep learning methods for predicting the sequence specificities of DBPs, they face two challenges as follows. Classic deep learning methods for DBPs prediction usually fail to capture the dependencies between genomic sequences since their commonly used one-hot codes are mutually orthogonal. Besides, these methods usually perform poorly when samples are inadequate. To address these two challenges, we developed a novel language model for mining DBPs using human genomic data and ChIP-seq datasets with decaying learning rates, named DNA Fine-tuned Language Model (DFLM). It can capture the dependencies between genome sequences based on the context of human genomic data and then fine-tune the features of DBPs tasks using different ChIP-seq datasets. First, we compared DFLM with the existing widely used methods on 69 datasets and we achieved excellent performance. Moreover, we conducted comparative experiments on complex DBPs and small datasets. The results show that DFLM still achieved a significant improvement. Finally, through visualization analysis of one-hot encoding and DFLM, we found that one-hot encoding completely cut off the dependencies of DNA sequences themselves, while DFLM using language models can well represent the dependency of DNA sequences. Source code are available at: https://github.com/Deep-Bioinfo/DFLM.

The Interactome of Protein, DNA, and RNA

Proteins participate in many processes of the organism and are very important for maintaining the health of the organism. However, proteins cannot function independently in the body. They must interact with proteins, DNA, RNA, and other substances to perform biological functions and maintain the body's health. At present, there are many experimental methods and software tools that can detect and predict the interaction between proteins and other substances. There are also many databases that record the interaction between proteins and other substances. This article mainly describes protein-protein, protein-DNA, and protein-RNA interactions in detail by introducing some commonly used experimental methods, the software tools produced with the accumulation of experimental data and the rapid development of machine learning, and the related databases that record the relationship between proteins and some substances. By this review, we hope that through the analysis and summary of various aspects, it will be convenient for researchers to conduct further research on protein interactions.

A Method for Predicting DNA Motif Length Based On Deep Learning

A DNA motif is a sequence pattern shared by the DNA sequence segments that bind to a specific protein. Discovering motifs in a given DNA sequence dataset plays a vital role in studying gene expression regulation. As an important attribute of the DNA motif, the motif length directly affects the quality of the discovered motifs. How to determine the motif length more accurately remains a difficult challenge to be solved. We propose a new motif length prediction scheme named MotifLen by using supervised machine learning. First, a method of constructing sample data for predicting the motif length is proposed. Secondly, a deep learning model for motif length prediction is constructed based on the convolutional neural network. Then, the methods of applying the proposed prediction model based on a motif found by an existing motif discovery algorithm are given. The experimental results show that i) the prediction accuracy of MotifLen is more than 90% on the validation set and is significantly higher than that of the compared methods on real datasets, ii) MotifLen can successfully optimize the motifs found by the existing motif discovery algorithms, and iii) it can effectively improve the time performance of some existing motif discovery algorithms.

Convolution Neural Networks Using Deep Matrix Factorization for Predicting Circrna-Disease Association

CircRNAs have a stable structure, which gives them a higher tolerance to nucleases. Therefore, the properties of circular RNAs are beneficial in disease diagnosis. However, there are few known associations between circRNAs and disease. Biological experiments identify new associations is time-consuming and high-cost. As a result, there is a need of building efficient and achievable computation models to predict potential circRNA-disease associations. In this paper, we design a novel convolution neural networks framework(DMFCNNCD) to learn features from deep matrix factorization to predict circRNA-disease associations. Firstly, we decompose the circRNA-disease association matrix to obtain the original features of the disease and circRNA, and use the mapping module to extract potential nonlinear features. Then, we integrate it with the similarity information to form a training set. Finally, we apply convolution neural networks to predict the unknown association between circRNAs and diseases. The five-fold cross-validation on various experiments shows that our method can predict circRNA-disease association and outperforms state of the art methods.

Tackling prediction uncertainty in machine learning for healthcare

Predictive machine-learning systems often do not convey the degree of confidence in the correctness of their outputs. To prevent unsafe prediction failures from machine-learning models, the users of the systems should be aware of the general accuracy of the model and understand the degree of confidence in each individual prediction. In this Perspective, we convey the need of prediction-uncertainty metrics in healthcare applications, with a focus on radiology. We outline the sources of prediction uncertainty, discuss how to implement prediction-uncertainty metrics in applications that require zero tolerance to errors and in applications that are error-tolerant, and provide a concise framework for understanding prediction uncertainty in healthcare contexts. For machine-learning-enabled automation to substantially impact healthcare, machine-learning models with zero tolerance for false-positive or false-negative errors must be developed intentionally.

Deep learning for MYC binding site recognition

Motivation: The definition of the genome distribution of the Myc transcription factor is extremely important since it may help predict its transcriptional activity particularly in the context of cancer. Myc is among the most powerful oncogenes involved in the occurrence and development of more than 80% of different types of pediatric and adult cancers. Myc regulates thousands of genes which can be in part different, depending on the type of tissues and tumours. Myc distribution along the genome has been determined experimentally through chromatin immunoprecipitation This approach, although powerful, is very time consuming and cannot be routinely applied to tumours of individual patients. Thus, it becomes of paramount importance to develop in silico tools that can effectively and rapidly predict its distribution on a given cell genome. New advanced computational tools (DeeperBind) can then be successfully employed to determine the function of Myc in a specific tumour, and may help to devise new directions and approaches to experiments first and personalized and more effective therapeutic treatments for a single patient later on. Results: The use of DeeperBind with DeepRAM on Colab platform (Google) can effectively predict the binding sites for the MYC factor with an accuracy above 0.96 AUC, when trained with multiple cell lines. The analysis of the filters in DeeperBind trained models shows, besides the consensus sequence CACGTG classically associated to the MYC factor, also the other consensus sequences G/C box or TGGGA, respectively bound by the SP1 and MIZ-1 transcription factors, which are known to mediate the MYC repressive response. Overall, our findings suggest a stronger synergy between the machine learning tools as DeeperBind and biological experiments, which may reduce the time consuming experiments by providing a direction to guide them.

DeepLncPro: an interpretable convolutional neural network model for identifying long non-coding RNA promoters

Long non-coding RNA (lncRNA) plays important roles in a series of biological processes. The transcription of lncRNA is regulated by its promoter. Hence, accurate identification of lncRNA promoter will be helpful to understand its regulatory mechanisms. Since experimental techniques remain time consuming for gnome-wide promoter identification, developing computational tools to identify promoters are necessary. However, only few computational methods have been proposed for lncRNA promoter prediction and their performances still have room to be improved. In the present work, a convolutional neural network based model, called DeepLncPro, was proposed to identify lncRNA promoters in human and mouse. Comparative results demonstrated that DeepLncPro was superior to both state-of-the-art machine learning methods and existing models for identifying lncRNA promoters. Furthermore, DeepLncPro has the ability to extract and analyze transcription factor binding motifs from lncRNAs, which made it become an interpretable model. These results indicate that the DeepLncPro can server as a powerful tool for identifying lncRNA promoters. An open-source tool for DeepLncPro was provided at https://github.com/zhangtian-yang/DeepLncPro.

PlantBind: an attention-based multi-label neural network for predicting plant transcription factor binding sites

Identification of transcription factor binding sites (TFBSs) is essential to understanding of gene regulation. Designing computational models for accurate prediction of TFBSs is crucial because it is not feasible to experimentally assay all transcription factors (TFs) in all sequenced eukaryotic genomes. Although many methods have been proposed for the identification of TFBSs in humans, methods designed for plants are comparatively underdeveloped. Here, we present PlantBind, a method for integrated prediction and interpretation of TFBSs based on DNA sequences and DNA shape profiles. Built on an attention-based multi-label deep learning framework, PlantBind not only simultaneously predicts the potential binding sites of 315 TFs, but also identifies the motifs bound by transcription factors. During the training process, this model revealed a strong similarity among TF family members with respect to target binding sequences. Trans-species prediction performance using four Zea mays TFs demonstrated the suitability of this model for transfer learning. Overall, this study provides an effective solution for identifying plant TFBSs, which will promote greater understanding of transcriptional regulatory mechanisms in plants.

Decoding degeneration: the implementation of machine learning for clinical detection of neurodegenerative disorders

Machine learning represents a growing subfield of artificial intelligence with much promise in the diagnosis, treatment, and tracking of complex conditions, including neurodegenerative disorders such as Alzheimer's and Parkinson's diseases. While no definitive methods of diagnosis or treatment exist for either disease, researchers have implemented machine learning algorithms with neuroimaging and motion-tracking technology to analyze pathologically relevant symptoms and biomarkers. Deep learning algorithms such as neural networks and complex combined architectures have proven capable of tracking disease-linked changes in brain structure and physiology as well as patient motor and cognitive symptoms and responses to treatment. However, such techniques require further development aimed at improving transparency, adaptability, and reproducibility. In this review, we provide an overview of existing neuroimaging technologies and supervised and unsupervised machine learning techniques with their current applications in the context of Alzheimer's and Parkinson's diseases.

Artificial Immune Cell, AI-cell, a New Tool to Predict Interferon Production by Peripheral Blood Monocytes in Response to Nucleic Acid Nanoparticles

Nucleic acid nanoparticles, or NANPs, rationally designed to communicate with the human immune system, can offer innovative therapeutic strategies to overcome the limitations of traditional nucleic acid therapies. Each set of NANPs is unique in their architectural parameters and physicochemical properties, which together with the type of delivery vehicles determine the kind and the magnitude of their immune response. Currently, there are no predictive tools that would reliably guide the design of NANPs to the desired immunological outcome, a step crucial for the success of personalized therapies. Through a systematic approach investigating physicochemical and immunological profiles of a comprehensive panel of various NANPs, the research team developes and experimentally validates a computational model based on the transformer architecture able to predict the immune activities of NANPs. It is anticipated that the freely accessible computational tool that is called an "artificial immune cell," or AI-cell, will aid in addressing the current critical public health challenges related to safety criteria of nucleic acid therapies in a timely manner and promote the development of novel biomedical tools.

DPB-NBFnet: Using neural Bellman-Ford networks to predict DNA-protein binding

DNA is a hereditary material that plays an essential role in micro-organisms and almost all other organisms. Meanwhile, proteins are a vital composition and principal undertaker of microbe movement. Therefore, studying the bindings between DNA and proteins is of high significance from the micro-biological point of view. In addition, the binding affinity prediction is beneficial for the study of drug design. However, existing experimental methods to identifying DNA-protein bindings are extremely expensive and time consuming. To solve this problem, many deep learning methods (including graph neural networks) have been developed to predict DNA-protein interactions. Our work possesses the same motivation and we put the latest Neural Bellman-Ford neural networks (NBFnets) into use to build pair representations of DNA and protein to predict the existence of DNA-protein binding (DPB). NBFnet is a graph neural network model that uses the Bellman-Ford algorithms to get pair representations and has been proven to have a state-of-the-art performance when used to solve the link prediction problem. After building the pair representations, we designed a feed-forward neural network structure and got a 2-D vector output as a predicted value of positive or negative samples. We conducted our experiments on 100 datasets from ENCODE datasets. Our experiments indicate that the performance of DPB-NBFnet is competitive when compared with the baseline models. We have also executed parameter tuning with different architectures to explore the structure of our framework.

Minimum Functional Length Analysis of K-Mer Based on BPNN

BP neural network (BPNN), as a multilayer feed-forward network, can realize the deep cognition to target data and high accuracy to output results. However, there were still no related research of k-mer based on BPNN yet. In present study, BPNN was used to train and test binary classification data of each classification mode respectively. All k-mer were divided into two categories according to the X + Y content or completely random mode. Results showed that 1) For classification mode of X + Y content, the accuracy of k-mers classification was 100 percent, no matter k ≤ 6 or k ≥ 7; 2) For completely random classification mode, the accuracy of classification is 100 percent for k-mers of k ≤ 6; But for k-mers of k ≥ 7, the accuracy is less than 100 percent, and with the increase of k value, the accuracy of classification gradually decreases (gradually approaches 50 percent). The k-mers of k ≥ 7 should be the basic functional fragment of nucleic acid, and perform basic nucleic acid function in the DNA sequence. The k-mers of k ≤ 6 should be the basic component fragment of nucleic acid, and no longer perform basic nucleic acid function.

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.

Prediction of Transcription Factor Binding Sites Using a Combined Deep Learning Approach

In the process of regulating gene expression and evolution, such as DNA replication and mRNA transcription, the binding of transcription factors (TFs) to TF binding sites (TFBS) plays a vital role. Precisely modeling the specificity of genes and searching for TFBS are helpful to explore the mechanism of cell expression. In recent years, computational and deep learning methods searching for TFBS have become an active field of research. However, existing methods generally cannot meet high performance and interpretability simultaneously. Here, we develop an accurate and interpretable attention-based hybrid approach, DeepARC, that combines a convolutional neural network (CNN) and recurrent neural network (RNN) to predict TFBS. DeepARC employs a positional embedding method to extract the hidden embedding from DNA sequences, including the positional information from OneHot encoding and the distributed embedding from DNA2Vec. DeepARC feeds the positional embedding of the DNA sequence into a CNN-BiLSTM-Attention-based framework to complete the task of finding the motif. Taking advantage of the attention mechanism, DeepARC can gain greater access to valuable information about the motif and bring interpretability to the work of searching for motifs through the attention weight graph. Moreover, DeepARC achieves promising performances with an average area under the receiver operating characteristic curve (AUC) score of 0.908 on five cell lines (A549, GM12878, Hep-G2, H1-hESC, and Hela) in the benchmark dataset. We also compare the positional embedding with OneHot and DNA2Vec and gain a competitive advantage.

A survey on protein–DNA-binding sites in computational biology

Transcription factors are important cellular components of the process of gene expression control. Transcription factor binding sites are locations where transcription factors specifically recognize DNA sequences, targeting gene-specific regions and recruiting transcription factors or chromatin regulators to fine-tune spatiotemporal gene regulation. As the common proteins, transcription factors play a meaningful role in life-related activities. In the face of the increase in the protein sequence, it is urgent how to predict the structure and function of the protein effectively. At present, protein–DNA-binding site prediction methods are based on traditional machine learning algorithms and deep learning algorithms. In the early stage, we usually used the development method based on traditional machine learning algorithm to predict protein–DNA-binding sites. In recent years, methods based on deep learning to predict protein–DNA-binding sites from sequence data have achieved remarkable success. Various statistical and machine learning methods used to predict the function of DNA-binding proteins have been proposed and continuously improved. Existing deep learning methods for predicting protein–DNA-binding sites can be roughly divided into three categories: convolutional neural network (CNN), recursive neural network (RNN) and hybrid neural network based on CNN–RNN. The purpose of this review is to provide an overview of the computational and experimental methods applied in the field of protein–DNA-binding site prediction today. This paper introduces the methods of traditional machine learning and deep learning in protein–DNA-binding site prediction from the aspects of data processing characteristics of existing learning frameworks and differences between basic learning model frameworks. Our existing methods are relatively simple compared with natural language processing, computational vision, computer graphics and other fields. Therefore, the summary of existing protein–DNA-binding site prediction methods will help researchers better understand this field.

Predicting residues involved in anti-DNA autoantibodies with limited neural networks

Abstract

Computer-aided rational vaccine design (RVD) and synthetic pharmacology are rapidly developing fields that leverage existing datasets for developing compounds of interest. Computational proteomics utilizes algorithms and models to probe proteins for functional prediction. A potentially strong target for computational approach is autoimmune antibodies, which are the result of broken tolerance in the immune system where it cannot distinguish “self” from “non-self” resulting in attack of its own structures (proteins and DNA, mainly). The information on structure, function, and pathogenicity of autoantibodies may assist in engineering RVD against autoimmune diseases. Current computational approaches exploit large datasets curated with extensive domain knowledge, most of which include the need for many resources and have been applied indirectly to problems of interest for DNA, RNA, and monomer protein binding. We present a novel method for discovering potential binding sites. We employed long short-term memory (LSTM) models trained on FASTA primary sequences to predict protein binding in DNA-binding hydrolytic antibodies (abzymes). We also employed CNN models applied to the same dataset for comparison with LSTM. While the CNN model outperformed the LSTM on the primary task of binding prediction, analysis of internal model representations of both models showed that the LSTM models recovered sub-sequences that were strongly correlated with sites known to be involved in binding. These results demonstrate that analysis of internal processes of LSTM models may serve as a powerful tool for primary sequence analysis.

Graphical abstract

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.

Towards a robust out-of-the-box neural network model for genomic data

Background

The accurate prediction of biological features from genomic data is paramount for precision medicine and sustainable agriculture. For decades, neural network models have been widely popular in fields like computer vision, astrophysics and targeted marketing given their prediction accuracy and their robust performance under big data settings. Yet neural network models have not made a successful transition into the medical and biological world due to the ubiquitous characteristics of biological data such as modest sample sizes, sparsity, and extreme heterogeneity.

Results

Here, we investigate the robustness, generalization potential and prediction accuracy of widely used convolutional neural network and natural language processing models with a variety of heterogeneous genomic datasets. Mainly, recurrent neural network models outperform convolutional neural network models in terms of prediction accuracy, overfitting and transferability across the datasets under study.

Conclusions

While the perspective of a robust out-of-the-box neural network model is out of reach, we identify certain model characteristics that translate well across datasets and could serve as a baseline model for translational researchers.

RNA editing increases the nucleotide diversity of SARS-CoV-2 in human host cells

SARS-CoV-2 is a positive-sense, single-stranded RNA virus responsible for the COVID-19 pandemic. It remains unclear whether and to what extent the virus in human host cells undergoes RNA editing, a major RNA modification mechanism. Here we perform a robust bioinformatic analysis of metatranscriptomic data from multiple bronchoalveolar lavage fluid samples of COVID-19 patients, revealing an appreciable number of A-to-I RNA editing candidate sites in SARS-CoV-2. We confirm the enrichment of A-to-I RNA editing signals at these candidate sites through evaluating four characteristics specific to RNA editing: the inferred RNA editing sites exhibit (i) stronger ADAR1 binding affinity predicted by a deep-learning model built from ADAR1 CLIP-seq data, (ii) decreased editing levels in ADAR1-inhibited human lung cells, (iii) local clustering patterns, and (iv) higher RNA secondary structure propensity. Our results have critical implications in understanding the evolution of SARS-CoV-2 as well as in COVID-19 research, such as phylogenetic analysis and vaccine development.

The COVID-19 pandemic is caused by SARS-CoV-2, an RNA virus. In the cells of COVID-19 patients, SARS-CoV-2 interacts with human proteins and is potentially subjected to their enzymatic activities. Here we investigated whether human protein enzymes can change the nucleotide sequence of SARS-CoV-2, thereby leaving a unique molecular footprint. We developed a robust computational algorithm to analyze the sequence data of SARS-CoV-2 obtained from lung fluid samples of COVID-19 patients and found that the virus contains new nucleotide changes that are likely induced by ADAR1, a powerful human protein that can modify specific nucleotide positions in many human transcripts. We further confirmed that the characteristics of the nucleotide changes detected in SARS-CoV-2 are similar to those observed in the human genes. Thus, these ADAR1-induced nucleotide changes may represent an under-appreciated force that can affect the evolution of SARS-CoV-2. Our study helps researchers better understand the evolutionary trajectory of SARS-CoV-2.

fastISM: performant in silico saturation mutagenesis for convolutional neural networks

MOTIVATION

Deep-learning models, such as convolutional neural networks, are able to accurately map biological sequences to associated functional readouts and properties by learning predictive de novo representations. In silico saturation mutagenesis (ISM) is a popular feature attribution technique for inferring contributions of all characters in an input sequence to the model's predicted output. The main drawback of ISM is its runtime, as it involves multiple forward propagations of all possible mutations of each character in the input sequence through the trained model to predict the effects on the output.

RESULTS

We present fastISM, an algorithm that speeds up ISM by a factor of over 10× for commonly used convolutional neural network architectures. fastISM is based on the observations that the majority of computation in ISM is spent in convolutional layers, and a single mutation only disrupts a limited region of intermediate layers, rendering most computation redundant. fastISM reduces the gap between backpropagation-based feature attribution methods and ISM. It far surpasses the runtime of backpropagation-based methods on multi-output architectures, making it feasible to run ISM on a large number of sequences.

AVAILABILITY AND IMPLEMENTATION

An easy-to-use Keras/TensorFlow 2 implementation of fastISM is available at https://github.com/kundajelab/fastISM. fastISM can be installed using pip install fastism. A hands-on tutorial can be found at https://colab.research.google.com/github/kundajelab/fastISM/blob/master/notebooks/colab/DeepSEA.ipynb.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Learning Useful Representations of DNA Sequences From ChIP-Seq Datasets for Exploring Transcription Factor Binding Specificities

Deep learning has been successfully applied to surprisingly different domains. Researchers and practitioners are employing trained deep learning models to enrich our knowledge. Transcription factors (TFs)are essential for regulating gene expression in all organisms by binding to specific DNA sequences. Here, we designed a deep learning model named SemanticCS (Semantic ChIP-seq)to predict TF binding specificities. We trained our learning model on an ensemble of ChIP-seq datasets (Multi-TF-cell)to learn useful intermediate features across multiple TFs and cells. To interpret these feature vectors, visualization analysis was used. Our results indicate that these learned representations can be used to train shallow machines for other tasks. Using diverse experimental data and evaluation metrics, we show that SemanticCS outperforms other popular methods. In addition, from experimental data, SemanticCS can help to identify the substitutions that cause regulatory abnormalities and to evaluate the effect of substitutions on the binding affinity for the RXR transcription factor. The online server for SemanticCS is freely available at http://qianglab.scst.suda.edu.cn/semanticCS/.

Base-resolution prediction of transcription factor binding signals by a deep learning framework

Transcription factors (TFs) play an important role in regulating gene expression, thus the identification of the sites bound by them has become a fundamental step for molecular and cellular biology. In this paper, we developed a deep learning framework leveraging existing fully convolutional neural networks (FCN) to predict TF-DNA binding signals at the base-resolution level (named as FCNsignal). The proposed FCNsignal can simultaneously achieve the following tasks: (i) modeling the base-resolution signals of binding regions; (ii) discriminating binding or non-binding regions; (iii) locating TF-DNA binding regions; (iv) predicting binding motifs. Besides, FCNsignal can also be used to predict opening regions across the whole genome. The experimental results on 53 TF ChIP-seq datasets and 6 chromatin accessibility ATAC-seq datasets show that our proposed framework outperforms some existing state-of-the-art methods. In addition, we explored to use the trained FCNsignal to locate all potential TF-DNA binding regions on a whole chromosome and predict DNA sequences of arbitrary length, and the results show that our framework can find most of the known binding regions and accept sequences of arbitrary length. Furthermore, we demonstrated the potential ability of our framework in discovering causal disease-associated single-nucleotide polymorphisms (SNPs) through a series of experiments.