iProDNA-CapsNet: identifying protein-DNA binding residues using capsule neural networks

Improving prediction performance of general protein language model by domain-adaptive pretraining on DNA-binding protein

DNA-protein interactions exert the fundamental structure of many pivotal biological processes, such as DNA replication, transcription, and gene regulation. However, accurate and efficient computational methods for identifying these interactions are still lacking. In this study, we propose a method ESM-DBP through refining the DNA-binding protein sequence repertory and domain-adaptive pretraining based the general protein language model. Our method considers the lacking exploration of general language model for DNA-binding protein domain-specific knowledge, so we screen out 170,264 DNA-binding protein sequences to construct the domain-adaptive language model. Experimental results on four downstream tasks show that ESM-DBP provides a better feature representation of DNA-binding protein compared to the original language model, resulting in improved prediction performance and outperforming the state-of-the-art methods. Moreover, ESM-DBP can still perform well even for those sequences with only a few homologous sequences. ChIP-seq on two predicted cases further support the validity of the proposed method.

A comprehensive review of protein-centric predictors for biomolecular interactions: from proteins to nucleic acids and beyond

Proteins interact with diverse ligands to perform a large number of biological functions, such as gene expression and signal transduction. Accurate identification of these protein-ligand interactions is crucial to the understanding of molecular mechanisms and the development of new drugs. However, traditional biological experiments are time-consuming and expensive. With the development of high-throughput technologies, an increasing amount of protein data is available. In the past decades, many computational methods have been developed to predict protein-ligand interactions. Here, we review a comprehensive set of over 160 protein-ligand interaction predictors, which cover protein-protein, protein-nucleic acid, protein-peptide and protein-other ligands (nucleotide, heme, ion) interactions. We have carried out a comprehensive analysis of the above four types of predictors from several significant perspectives, including their inputs, feature profiles, models, availability, etc. The current methods primarily rely on protein sequences, especially utilizing evolutionary information. The significant improvement in predictions is attributed to deep learning methods. Additionally, sequence-based pretrained models and structure-based approaches are emerging as new trends.

i5mC-DCGA: an improved hybrid network framework based on the CBAM attention mechanism for identifying promoter 5mC sites

BACKGROUND

5-Methylcytosine (5mC) plays a very important role in gene stability, transcription, and development. Therefore, accurate identification of the 5mC site is of key importance in genetic and pathological studies. However, traditional experimental methods for identifying 5mC sites are time-consuming and costly, so there is an urgent need to develop computational methods to automatically detect and identify these 5mC sites.

RESULTS

Deep learning methods have shown great potential in the field of 5mC sites, so we developed a deep learning combinatorial model called i5mC-DCGA. The model innovatively uses the Convolutional Block Attention Module (CBAM) to improve the Dense Convolutional Network (DenseNet), which is improved to extract advanced local feature information. Subsequently, we combined a Bidirectional Gated Recurrent Unit (BiGRU) and a Self-Attention mechanism to extract global feature information. Our model can learn feature representations of abstract and complex from simple sequence coding, while having the ability to solve the sample imbalance problem in benchmark datasets. The experimental results show that the i5mC-DCGA model achieves 97.02%, 96.52%, 96.58% and 85.58% in sensitivity (Sn), specificity (Sp), accuracy (Acc) and matthews correlation coefficient (MCC), respectively.

CONCLUSIONS

The i5mC-DCGA model outperforms other existing prediction tools in predicting 5mC sites, and it is currently the most representative promoter 5mC site prediction tool. The benchmark dataset and source code for the i5mC-DCGA model can be found in https://github.com/leirufeng/i5mC-DCGA .

ULDNA: integrating unsupervised multi-source language models with LSTM-attention network for high-accuracy protein-DNA binding site prediction

Efficient and accurate recognition of protein-DNA interactions is vital for understanding the molecular mechanisms of related biological processes and further guiding drug discovery. Although the current experimental protocols are the most precise way to determine protein-DNA binding sites, they tend to be labor-intensive and time-consuming. There is an immediate need to design efficient computational approaches for predicting DNA-binding sites. Here, we proposed ULDNA, a new deep-learning model, to deduce DNA-binding sites from protein sequences. This model leverages an LSTM-attention architecture, embedded with three unsupervised language models that are pre-trained on large-scale sequences from multiple database sources. To prove its effectiveness, ULDNA was tested on 229 protein chains with experimental annotation of DNA-binding sites. Results from computational experiments revealed that ULDNA significantly improves the accuracy of DNA-binding site prediction in comparison with 17 state-of-the-art methods. In-depth data analyses showed that the major strength of ULDNA stems from employing three transformer language models. Specifically, these language models capture complementary feature embeddings with evolution diversity, in which the complex DNA-binding patterns are buried. Meanwhile, the specially crafted LSTM-attention network effectively decodes evolution diversity-based embeddings as DNA-binding results at the residue level. Our findings demonstrated a new pipeline for predicting DNA-binding sites on a large scale with high accuracy from protein sequence alone.

BiCaps-DBP: Predicting DNA-binding proteins from protein sequences using Bi-LSTM and a 1D-capsule network

Predicting DNA-binding proteins (DBPs) based solely on primary sequences is one of the most challenging problems in genome annotation. DBPs play a crucial role in various biological processes, including DNA replication, transcription, repair, and splicing. Some DBPs are essential in pharmaceutical research on various human cancers and autoimmune diseases. Existing experimental methods for identifying DBPs are time-consuming and costly. Therefore, developing a rapid and accurate computational technique is necessary to address the issue. This study introduces BiCaps-DBP, a deep learning-based method that improves DBP prediction performance by combining bidirectional long short-term memory with a 1D-capsule network. This study uses three training and independent datasets to evaluate the proposed model's generalizability and robustness. Based on three independent datasets, BiCaps-DBP achieved 1.05%, 5.79% and 0.40% higher accuracies than an existing predictor for PDB2272, PDB186 and PDB20000, respectively. These outcomes indicate that the proposed method is a promising DBP predictor.

i4mC-GRU: Identifying DNA N4-Methylcytosine sites in mouse genomes using bidirectional gated recurrent unit and sequence-embedded features

N⁴-methylcytosine (4mC) is one of the most common DNA methylation modifications found in both prokaryotic and eukaryotic genomes. Since the 4mC has various essential biological roles, determining its location helps reveal unexplored physiological and pathological pathways. In this study, we propose an effective computational method called i4mC-GRU using a gated recurrent unit and duplet sequence-embedded features to predict potential 4mC sites in mouse (Mus musculus) genomes. To fairly assess the performance of the model, we compared our method with several state-of-the-art methods using two different benchmark datasets. Our results showed that i4mC-GRU achieved area under the receiver operating characteristic curve values of 0.97 and 0.89 and area under the precision-recall curve values of 0.98 and 0.90 on the first and second benchmark datasets, respectively. Briefly, our method outperformed existing methods in predicting 4mC sites in mouse genomes. Also, we deployed i4mC-GRU as an online web server, supporting users in genomics studies.

CapsNet-MHC predicts peptide-MHC class I binding based on capsule neural networks

The Major Histocompatibility Complex (MHC) binds to the derived peptides from pathogens to present them to killer T cells on the cell surface. Developing computational methods for accurate, fast, and explainable peptide-MHC binding prediction can facilitate immunotherapies and vaccine development. Various deep learning-based methods rely on separate feature extraction from the peptide and MHC sequences and ignore their pairwise binding information. This paper develops a capsule neural network-based method to efficiently capture the peptide-MHC complex features to predict the peptide-MHC class I binding. Various evaluations confirmed our method outperformance over the alternative methods, while it can provide accurate prediction over less available data. Moreover, for providing precise insights into the results, we explored the essential features that contributed to the prediction. Since the simulation results demonstrated consistency with the experimental studies, we concluded that our method can be utilized for the accurate, rapid, and interpretable peptide-MHC binding prediction to assist biological therapies.

Protein-DNA Binding Residue Prediction via Bagging Strategy and Sequence-Based Cube-Format Feature

Protein-DNA interactions play an important role in diverse biological processes. Accurately identifying protein-DNA binding residues is a critical but challenging task for protein function annotations and drug design. Although wet-lab experimental methods are the most accurate way to identify protein-DNA binding residues, they are time consuming and labor intensive. There is an urgent need to develop computational methods to rapidly and accurately predict protein-DNA binding residues. In this study, we propose a novel sequence-based method, named PredDBR, for predicting DNA-binding residues. In PredDBR, for each query protein, its position-specific frequency matrix (PSFM), predicted secondary structure (PSS), and predicted probabilities of ligand-binding residues (PPLBR) are first generated as three feature sources. Secondly, for each feature source, the sliding window technique is employed to extract the matrix-format feature of each residue. Then, we design two strategies, i.e., square root (SR) and average (AVE), to separately transform PSFM-based and two predicted feature source-based, i.e., PSS-based and PPLBR-based, matrix-format features of each residue into three corresponding cube-format features. Finally, after serially combining the three cube-format features, the ensemble classifier is generated via applying bagging strategy to multiple base classifiers built by the framework of 2D convolutional neural network. The computational experimental results demonstrate that the proposed PredDBR achieves an average overall accuracy of 93.7% and a Mathew's correlation coefficient of 0.405 on two independent validation datasets and outperforms several state-of-the-art sequenced-based protein-DNA binding residue predictors. The PredDBR web-server is available at https://jun-csbio.github.io/PredDBR/.

iPromoter-Seqvec: identifying promoters using bidirectional long short-term memory and sequence-embedded features

BACKGROUND

Promoters, non-coding DNA sequences located at upstream regions of the transcription start site of genes/gene clusters, are essential regulatory elements for the initiation and regulation of transcriptional processes. Furthermore, identifying promoters in DNA sequences and genomes significantly contributes to discovering entire structures of genes of interest. Therefore, exploration of promoter regions is one of the most imperative topics in molecular genetics and biology. Besides experimental techniques, computational methods have been developed to predict promoters. In this study, we propose iPromoter-Seqvec - an efficient computational model to predict TATA and non-TATA promoters in human and mouse genomes using bidirectional long short-term memory neural networks in combination with sequence-embedded features extracted from input sequences. The promoter and non-promoter sequences were retrieved from the Eukaryotic Promoter database and then were refined to create four benchmark datasets.

RESULTS

The area under the receiver operating characteristic curve (AUCROC) and the area under the precision-recall curve (AUCPR) were used as two key metrics to evaluate model performance. Results on independent test sets showed that iPromoter-Seqvec outperformed other state-of-the-art methods with AUCROC values ranging from 0.85 to 0.99 and AUCPR values ranging from 0.86 to 0.99. Models predicting TATA promoters in both species had slightly higher predictive power compared to those predicting non-TATA promoters. With a novel idea of constructing artificial non-promoter sequences based on promoter sequences, our models were able to learn highly specific characteristics discriminating promoters from non-promoters to improve predictive efficiency.

CONCLUSIONS

iPromoter-Seqvec is a stable and robust model for predicting both TATA and non-TATA promoters in human and mouse genomes. Our proposed method was also deployed as an online web server with a user-friendly interface to support research communities. Links to our source codes and web server are available at https://github.com/mldlproject/2022-iPromoter-Seqvec .

Identifying Transcription Factors That Prefer Binding to Methylated DNA Using Reduced G-Gap Dipeptide Composition

Transcription factors (TFs) play an important role in gene expression and regulation of 3D genome conformation. TFs have ability to bind to specific DNA fragments called enhancers and promoters. Some TFs bind to promoter DNA fragments which are near the transcription initiation site and form complexes that allow polymerase enzymes to bind to initiate transcription. Previous studies showed that methylated DNAs had ability to inhibit and prevent TFs from binding to DNA fragments. However, recent studies have found that there were TFs that could bind to methylated DNA fragments. The identification of these TFs is an important steppingstone to a better understanding of cellular gene expression mechanisms. However, as experimental methods are often time-consuming and labor-intensive, developing computational methods is essential. In this study, we propose two machine learning methods for two problems: (1) identifying TFs and (2) identifying TFs that prefer binding to methylated DNA targets (TFPMs). For the TF identification problem, the proposed method uses the position-specific scoring matrix for data representation and a deep convolutional neural network for modeling. This method achieved 90.56% sensitivity, 83.96% specificity, and an area under the receiver operating characteristic curve (AUC) of 0.9596 on an independent test set. For the TFPM identification problem, we propose to use the reduced g-gap dipeptide composition for data representation and the support vector machine algorithm for modeling. This method achieved 82.61% sensitivity, 64.86% specificity, and an AUC of 0.8486 on another independent test set. These results are higher than those of other studies on the same problems.

A deep learning-based method for the prediction of DNA interacting residues in a protein

DNA-protein interaction is one of the most crucial interactions in the biological system, which decides the fate of many processes such as transcription, regulation and splicing of genes. In this study, we trained our models on a training dataset of 646 DNA-binding proteins having 15 636 DNA interacting and 298 503 non-interacting residues. Our trained models were evaluated on an independent dataset of 46 DNA-binding proteins having 965 DNA interacting and 9911 non-interacting residues. All proteins in the independent dataset have less than 30% of sequence similarity with proteins in the training dataset. A wide range of traditional machine learning and deep learning (1D-CNN) techniques-based models have been developed using binary, physicochemical properties and Position-Specific Scoring Matrix (PSSM)/evolutionary profiles. In the case of machine learning technique, eXtreme Gradient Boosting-based model achieved a maximum area under the receiver operating characteristics (AUROC) curve of 0.77 on the independent dataset using PSSM profile. Deep learning-based model achieved the highest AUROC of 0.79 on the independent dataset using a combination of all three profiles. We evaluated the performance of existing methods on the independent dataset and observed that our proposed method outperformed all the existing methods. In order to facilitate scientific community, we developed standalone software and web server, which are accessible from https://webs.iiitd.edu.in/raghava/dbpred.

Inadequacy of evolutionary profiles vis-a-vis single sequences in predicting transient DNA-binding sites in proteins

Sequence-based prediction of DNA-binding residues in a protein is a widely studied problem for which machine learning methods with continuously improving predictive power have been developed. Concatenated rows within a sliding window of a Position Specific Substitution Matrix (PSSM) of the protein concerned are currently used as the primary feature set in almost all the methods of predicting DNA-binding residues. Here we report that these evolutionary profiles are powerful, only for identifying conserved binding sites and fall short for the residue positions which undergo binding to non-binding transitions in closely related proteins. We created a database of highly similar protein pairs with known protein-DNA complexes and investigated differential predictability of conserved and transient binding within each pair. Retraining machine learning models uniformly, we compared the predictive powers of the models trained on PSSMs against similarly trained models on sparse-encoded single sequences. We found that the transient binding site predictions from evolutionary profiles are outperformed by single sequence based models under controlled training and test experiments by as much as 8 percentage points. Thus, we conclude that the PSSM-based models are inadequate to predict high specificity DNA-binding residues. These findings are of critical significance for the design of mutant- and species-specific DNA ligands and for homology based modeling of protein-DNA complexes.

DeepDISOBind: accurate prediction of RNA-, DNA- and protein-binding intrinsically disordered residues with deep multi-task learning

Proteins with intrinsically disordered regions (IDRs) are common among eukaryotes. Many IDRs interact with nucleic acids and proteins. Annotation of these interactions is supported by computational predictors, but to date, only one tool that predicts interactions with nucleic acids was released, and recent assessments demonstrate that current predictors offer modest levels of accuracy. We have developed DeepDISOBind, an innovative deep multi-task architecture that accurately predicts deoxyribonucleic acid (DNA)-, ribonucleic acid (RNA)- and protein-binding IDRs from protein sequences. DeepDISOBind relies on an information-rich sequence profile that is processed by an innovative multi-task deep neural network, where subsequent layers are gradually specialized to predict interactions with specific partner types. The common input layer links to a layer that differentiates protein- and nucleic acid-binding, which further links to layers that discriminate between DNA and RNA interactions. Empirical tests show that this multi-task design provides statistically significant gains in predictive quality across the three partner types when compared to a single-task design and a representative selection of the existing methods that cover both disorder- and structure-trained tools. Analysis of the predictions on the human proteome reveals that DeepDISOBind predictions can be encoded into protein-level propensities that accurately predict DNA- and RNA-binding proteins and protein hubs. DeepDISOBind is available at https://www.csuligroup.com/DeepDISOBind/.

iCYP-MFE: Identifying Human Cytochrome P450 Inhibitors Using Multitask Learning and Molecular Fingerprint-Embedded Encoding

The human cytochrome P450 (CYP) superfamily holds responsibilities for the metabolism of both endogenous and exogenous compounds such as drugs, cellular metabolites, and toxins. The inhibition exerted on the CYP enzymes is closely associated with adverse drug reactions encompassing metabolic failures and induced side effects. In modern drug discovery, identification of potential CYP inhibitors is, therefore, highly essential. Alongside experimental approaches, numerous computational models have been proposed to address this biochemical issue. In this study, we introduce iCYP-MFE, a computational framework for virtual screening on CYP inhibitors toward 1A2, 2C9, 2C19, 2D6, and 3A4 isoforms. iCYP-MFE contains a set of five robust, stable, and effective prediction models developed using multitask learning incorporated with molecular fingerprint-embedded features. The results show that multitask learning can remarkably leverage useful information from related tasks to promote global performance. Comparative analysis indicates that iCYP-MFE achieves three predominant tasks, one equivalent task, and one less effective task compared to state-of-the-art methods. The area under the receiver operating characteristic curve (AUC-ROC) and the area under the precision-recall curve (AUC-PR) were two decisive metrics used for model evaluation. The prediction task for CYP2D6-inhibition achieves the highest AUC-ROC value of 0.93 while the prediction task for CYP1A2-inhibition obtains the highest AUC-PR value of 0.92. The substructural analysis preliminarily explains the nature of the CYP-inhibitory activity of compounds. An online web server for iCYP-MFE with a user-friendly interface was also deployed to support scientific communities in identifying CYP inhibitors.

Computational approaches to predict protein functional families and functional sites

Understanding the mechanisms of protein function is indispensable for many biological applications, such as protein engineering and drug design. However, experimental annotations are sparse, and therefore, theoretical strategies are needed to fill the gap. Here, we present the latest developments in building functional subclassifications of protein superfamilies and using evolutionary conservation to detect functional determinants, for example, catalytic-, binding- and specificity-determining residues important for delineating the functional families. We also briefly review other features exploited for functional site detection and new machine learning strategies for combining multiple features.

MDCAN-Lys: A Model for Predicting Succinylation Sites Based on Multilane Dense Convolutional Attention Network

Lysine succinylation is an important post-translational modification, whose abnormalities are closely related to the occurrence and development of many diseases. Therefore, exploring effective methods to identify succinylation sites is helpful for disease treatment and research of related drugs. However, most existing computational methods for the prediction of succinylation sites are still based on machine learning. With the increasing volume of data and complexity of feature representations, it is necessary to explore effective deep learning methods to recognize succinylation sites. In this paper, we propose a multilane dense convolutional attention network, MDCAN-Lys. MDCAN-Lys extracts sequence information, physicochemical properties of amino acids, and structural properties of proteins using a three-way network, and it constructs feature space. For each sub-network, MDCAN-Lys uses the cascading model of dense convolutional block and convolutional block attention module to capture feature information at different levels and improve the abstraction ability of the network. The experimental results of 10-fold cross-validation and independent testing show that MDCAN-Lys can recognize more succinylation sites, which is consistent with the conclusion of the case study. Thus, it is worthwhile to explore deep learning-based methods for the recognition of succinylation sites.

DeepD2V: A Novel Deep Learning-Based Framework for Predicting Transcription Factor Binding Sites from Combined DNA Sequence

Predicting in vivo protein-DNA binding sites is a challenging but pressing task in a variety of fields like drug design and development. Most promoters contain a number of transcription factor (TF) binding sites, but only a small minority has been identified by biochemical experiments that are time-consuming and laborious. To tackle this challenge, many computational methods have been proposed to predict TF binding sites from DNA sequence. Although previous methods have achieved remarkable performance in the prediction of protein-DNA interactions, there is still considerable room for improvement. In this paper, we present a hybrid deep learning framework, termed DeepD2V, for transcription factor binding sites prediction. First, we construct the input matrix with an original DNA sequence and its three kinds of variant sequences, including its inverse, complementary, and complementary inverse sequence. A sliding window of size k with a specific stride is used to obtain its k-mer representation of input sequences. Next, we use word2vec to obtain a pre-trained k-mer word distributed representation model. Finally, the probability of protein-DNA binding is predicted by using the recurrent and convolutional neural network. The experiment results on 50 public ChIP-seq benchmark datasets demonstrate the superior performance and robustness of DeepD2V. Moreover, we verify that the performance of DeepD2V using word2vec-based k-mer distributed representation is better than one-hot encoding, and the integrated framework of both convolutional neural network (CNN) and bidirectional LSTM (bi-LSTM) outperforms CNN or the bi-LSTM model when used alone. The source code of DeepD2V is available at the github repository.

TargetDBP+: Enhancing the Performance of Identifying DNA-Binding Proteins via Weighted Convolutional Features

Protein-DNA interactions exist ubiquitously and play important roles in the life cycles of living cells. The accurate identification of DNA-binding proteins (DBPs) is one of the key steps to understand the mechanisms of protein-DNA interactions. Although many DBP identification methods have been proposed, the current performance is still unsatisfactory. In this study, a new method, called TargetDBP+, is developed to further enhance the performance of identifying DBPs. In TargetDBP+, five convolutional features are first extracted from five feature sources, i.e., amino acid one-hot matrix (AAOHM), position-specific scoring matrix (PSSM), predicted secondary structure probability matrix (PSSPM), predicted solvent accessibility probability matrix (PSAPM), and predicted probabilities of DNA-binding sites (PPDBSs); second, the five features are weightedly and serially combined using the weights of all of the elements learned by the differential evolution algorithm; and finally, the DBP identification model of TargetDBP+ is trained using the support vector machine (SVM) algorithm. To evaluate the developed TargetDBP+ and compare it with other existing methods, a new gold-standard benchmark data set, called UniSwiss, is constructed, which consists of 4881 DBPs and 4881 non-DBPs extracted from the UniprotKB/Swiss-Prot database. Experimental results demonstrate that TargetDBP+ can obtain an accuracy of 85.83% and precision of 88.45% covering 82.41% of all DBP data on the independent validation subset of UniSwiss, with the MCC value (0.718) being significantly higher than those of other state-of-the-art control methods. The web server of TargetDBP+ is accessible at http://csbio.njust.edu.cn/bioinf/targetdbpplus/; the UniSwiss data set and stand-alone program of TargetDBP+ are accessible at https://github.com/jun-csbio/TargetDBPplus.

Identifying the sequence specificities of circRNA-binding proteins based on a capsule network architecture

Background

Circular RNAs (circRNAs) are widely expressed in cells and tissues and are involved in biological processes and human diseases. Recent studies have demonstrated that circRNAs can interact with RNA-binding proteins (RBPs), which is considered an important aspect for investigating the function of circRNAs.

Results

In this study, we design a slight variant of the capsule network, called circRB, to identify the sequence specificities of circRNAs binding to RBPs. In this model, the sequence features of circRNAs are extracted by convolution operations, and then, two dynamic routing algorithms in a capsule network are employed to discriminate between different binding sites by analysing the convolution features of binding sites. The experimental results show that the circRB method outperforms the existing computational methods. Afterwards, the trained models are applied to detect the sequence motifs on the seven circRNA-RBP bound sequence datasets and matched to known human RNA motifs. Some motifs on circular RNAs overlap with those on linear RNAs. Finally, we also predict binding sites on the reported full-length sequences of circRNAs interacting with RBPs, attempting to assist current studies. We hope that our model will contribute to better understanding the mechanisms of the interactions between RBPs and circRNAs.

Conclusion

In view of the poor studies about the sequence specificities of circRNA-binding proteins, we designed a classification framework called circRB based on the capsule network. The results show that the circRB method is an effective method, and it achieves higher prediction accuracy than other methods.