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

TransC-ac4C: Identification of N4-Acetylcytidine (ac4C) Sites in mRNA Using Deep Learning

N4-acetylcytidine (ac4C) is a post-transcriptional modification in mRNA that is critical in mRNA translation in terms of stability and regulation. In the past few years, numerous approaches employing convolutional neural networks (CNN) and Transformer have been proposed for the identification of ac4C sites, with each variety of approaches processing distinct characteristics. CNN-based methods excel at extracting local features and positional information, whereas Transformer-based ones stands out in establishing long-range dependencies and generating global representations. Given the importance of both local and global features in mRNA ac4C sites identification, we propose a novel method termed TransC-ac4C which combines CNN and Transformer together for enhancing the feature extraction capability and improving the identification accuracy. Five different feature encoding strategies (One-hot, NCP, ND, EIIP, and K-mer) are employed to generate the mRNA sequence representations, in which way the sequence attributes and physical and chemical properties of the sequences can be embedded. To strengthen the relevance of features, we construct a novel feature fusion method. Firstly, the CNN is employed to process five single features, stitch them together and feed them to the Transformer layer. Then, our approach employs CNN to extract local features and Transformer subsequently to establish global long-range dependencies among extracted features. We use 5-fold cross-validation to evaluate the model, and the evaluation indicators are significantly improved. The prediction accuracy of the two datasets is as high as 81.42% and 80.69%, respectively. It demonstrates the stronger competitiveness and generalization performance of our model.

An Efficient Exact Algorithm for Planted Motif Search on Large DNA Sequence Datasets

DNA motif is the pattern shared by similar fragments in DNA sequences, which plays a key role in regulating gene expression, and DNA motif discovery has become a key research topic. Exact planted ( l, d )-motif search (PMS) is one of the motif discovery approaches, which aims to find from t sequences all the ( l, d )-motifs that are motifs of l length appearing in at least qt sequences with at most d mismatches. The existing exact PMS algorithms are only suitable for small datasets of DNA sequences. The development of high-throughput sequencing technology generates vast amount of DNA sequence data, which brings challenges to solving exact PMS problems efficiently. Therefore, we propose an efficient exact PMS algorithm called PMmotif for large datasets of DNA sequences, after analyzing the time complexity of the existing exact PMS algorithms. PMmotif finds ( l, d )-motifs with strategy by searching the branches on the pattern tree that may contain ( l, d )-motifs. It is verified by experiments that the running time ratio of some existing excellent PMS algorithms to PMmotif is between 14.83 and 58.94. In addition, for the first time, PMmotif can solve the ( 15,5 )and ( 17,6 ) challenge problem instances on large DNA sequence datasets (3000 sequences of length 200) within 24 hours.

Proformer: a hybrid macaron transformer model predicts expression values from promoter sequences

The breakthrough high-throughput measurement of the cis-regulatory activity of millions of randomly generated promoters provides an unprecedented opportunity to systematically decode the cis-regulatory logic that determines the expression values. We developed an end-to-end transformer encoder architecture named Proformer to predict the expression values from DNA sequences. Proformer used a Macaron-like Transformer encoder architecture, where two half-step feed forward (FFN) layers were placed at the beginning and the end of each encoder block, and a separable 1D convolution layer was inserted after the first FFN layer and in front of the multi-head attention layer. The sliding k-mers from one-hot encoded sequences were mapped onto a continuous embedding, combined with the learned positional embedding and strand embedding (forward strand vs. reverse complemented strand) as the sequence input. Moreover, Proformer introduced multiple expression heads with mask filling to prevent the transformer models from collapsing when training on relatively small amount of data. We empirically determined that this design had significantly better performance than the conventional design such as using the global pooling layer as the output layer for the regression task. These analyses support the notion that Proformer provides a novel method of learning and enhances our understanding of how cis-regulatory sequences determine the expression values.

MetaAc4C: A multi-module deep learning framework for accurate prediction of N4-acetylcytidine sites based on pre-trained bidirectional encoder representation and generative adversarial networks

MOTIVATION

N4-acetylcytidine (ac4C) is a highly conserved RNA modification that plays a crucial role in various biological processes. Accurately identifying ac4C sites is of paramount importance for gaining a deeper understanding of their regulatory mechanisms. Nevertheless, the existing experimental techniques for ac4C site identification are characterized by limitations in terms of cost-effectiveness, while the performance of current computational methods in accurately identifying ac4C sites requires further enhancement.

RESULTS

In this paper, we present MetaAc4C, an advanced deep learning model that leverages pre-trained bidirectional encoder representations from transformers (BERT). The model is based on a bi-directional long short-term memory network (BLSTM) architecture, incorporating attention mechanism and residual connection. To address the issue of data imbalance, we adapt generative adversarial networks to generate synthetic feature samples. On the independent test set, MetaAc4C surpasses the current state-of-the-art ac4C prediction model, exhibiting improvements in terms of ACC, MCC, and AUROC by 2.36%, 4.76%, and 3.11%, respectively, on the unbalanced dataset. When evaluated on the balanced dataset, MetaAc4C achieves improvements in ACC, MCC, and AUROC by 2.6%, 5.11%, and 1.01%, respectively. Notably, our approach of utilizing WGAN-GP augmented training RNA samples demonstrates even superior performance compared to the SMOTE oversampling method.

iAVPs-ResBi: Identifying antiviral peptides by using deep residual network and bidirectional gated recurrent unit

Human history is also the history of the fight against viral diseases. From the eradication of viruses to coexistence, advances in biomedicine have led to a more objective understanding of viruses and a corresponding increase in the tools and methods to combat them. More recently, antiviral peptides (AVPs) have been discovered, which due to their superior advantages, have achieved great impact as antiviral drugs. Therefore, it is very necessary to develop a prediction model to accurately identify AVPs. In this paper, we develop the iAVPs-ResBi model using k-spaced amino acid pairs (KSAAP), encoding based on grouped weight (EBGW), enhanced grouped amino acid composition (EGAAC) based on the N5C5 sequence, composition, transition and distribution (CTD) based on physicochemical properties for multi-feature extraction. Then we adopt bidirectional long short-term memory (BiLSTM) to fuse features for obtaining the most differentiated information from multiple original feature sets. Finally, the deep model is built by combining improved residual network and bidirectional gated recurrent unit (BiGRU) to perform classification. The results obtained are better than those of the existing methods, and the accuracies are 95.07, 98.07, 94.29 and 97.50% on the four datasets, which show that iAVPs-ResBi can be used as an effective tool for the identification of antiviral peptides. The datasets and codes are freely available at https://github.com/yunyunliang88/iAVPs-ResBi.

Stacking-ac4C: an ensemble model using mixed features for identifying n4-acetylcytidine in mRNA

N4-acetylcytidine (ac4C) is a modification of cytidine at the nitrogen-4 position, playing a significant role in the translation process of mRNA. However, the precise mechanism and details of how ac4C modifies translated mRNA remain unclear. Since identifying ac4C sites using conventional experimental methods is both labor-intensive and time-consuming, there is an urgent need for a method that can promptly recognize ac4C sites. In this paper, we propose a comprehensive ensemble learning model, the Stacking-based heterogeneous integrated ac4C model, engineered explicitly to identify ac4C sites. This innovative model integrates three distinct feature extraction methodologies: Kmer, electron-ion interaction pseudo-potential values (PseEIIP), and pseudo-K-tuple nucleotide composition (PseKNC). The model also incorporates the robust Cluster Centroids algorithm to enhance its performance in dealing with imbalanced data and alleviate underfitting issues. Our independent testing experiments indicate that our proposed model improves the Mcc by 15.61% and the ROC by 5.97% compared to existing models. To test our model's adaptability, we also utilized a balanced dataset assembled by the authors of iRNA-ac4C. Our model showed an increase in Sn of 4.1%, an increase in Acc of nearly 1%, and ROC improvement of 0.35% on this balanced dataset. The code for our model is freely accessible at https://github.com/louliliang/ST-ac4C.git, allowing users to quickly build their model without dealing with complicated mathematical equations.

Early detection of hepatocellular carcinoma via no end-repair enzymatic methylation sequencing of cell-free DNA and pre-trained neural network

BACKGROUND

Early detection of hepatocellular carcinoma (HCC) is important in order to improve patient prognosis and survival rate. Methylation sequencing combined with neural networks to identify cell-free DNA (cfDNA) carrying aberrant methylation offers an appealing and non-invasive approach for HCC detection. However, some limitations exist in traditional methylation detection technologies and models, which may impede their performance in the read-level detection of HCC.

METHODS

We developed a low DNA damage and high-fidelity methylation detection method called No End-repair Enzymatic Methyl-seq (NEEM-seq). We further developed a read-level neural detection model called DeepTrace that can better identify HCC-derived sequencing reads through a pre-trained and fine-tuned neural network. After pre-training on 11 million reads from NEEM-seq, DeepTrace was fine-tuned using 1.2 million HCC-derived reads from tumor tissue DNA after noise reduction, and 2.7 million non-tumor reads from non-tumor cfDNA. We validated the model using data from 130 individuals with cfDNA whole-genome NEEM-seq at around 1.6X depth.

RESULTS

NEEM-seq overcomes the drawbacks of traditional enzymatic methylation sequencing methods by avoiding the introduction of unmethylation errors in cfDNA. DeepTrace outperformed other models in identifying HCC-derived reads and detecting HCC individuals. Based on the whole-genome NEEM-seq data of cfDNA, our model showed high accuracy of 96.2%, sensitivity of 93.6%, and specificity of 98.5% in the validation cohort consisting of 62 HCC patients, 48 liver disease patients, and 20 healthy individuals. In the early stage of HCC (BCLC 0/A and TNM I), the sensitivity of DeepTrace was 89.6 and 89.5% respectively, outperforming Alpha Fetoprotein (AFP) which showed much lower sensitivity in both BCLC 0/A (50.5%) and TNM I (44.7%).

CONCLUSIONS

By combining high-fidelity methylation data from NEEM-seq with the DeepTrace model, our method has great potential for HCC early detection with high sensitivity and specificity, making it potentially suitable for clinical applications. DeepTrace: https://github.com/Bamrock/DeepTrace.

Deep Learning for Genomics: From Early Neural Nets to Modern Large Language Models

The data explosion driven by advancements in genomic research, such as high-throughput sequencing techniques, is constantly challenging conventional methods used in genomics. In parallel with the urgent demand for robust algorithms, deep learning has succeeded in various fields such as vision, speech, and text processing. Yet genomics entails unique challenges to deep learning, since we expect a superhuman intelligence that explores beyond our knowledge to interpret the genome from deep learning. A powerful deep learning model should rely on the insightful utilization of task-specific knowledge. In this paper, we briefly discuss the strengths of different deep learning models from a genomic perspective so as to fit each particular task with proper deep learning-based architecture, and we remark on practical considerations of developing deep learning architectures for genomics. We also provide a concise review of deep learning applications in various aspects of genomic research and point out current challenges and potential research directions for future genomics applications. We believe the collaborative use of ever-growing diverse data and the fast iteration of deep learning models will continue to contribute to the future of genomics.

Predictive analyses of regulatory sequences with EUGENe

Deep learning has become a popular tool to study cis-regulatory function. Yet efforts to design software for deep-learning analyses in regulatory genomics that are findable, accessible, interoperable and reusable (FAIR) have fallen short of fully meeting these criteria. Here we present elucidating the utility of genomic elements with neural nets (EUGENe), a FAIR toolkit for the analysis of genomic sequences with deep learning. EUGENe consists of a set of modules and subpackages for executing the key functionality of a genomics deep learning workflow: (1) extracting, transforming and loading sequence data from many common file formats; (2) instantiating, initializing and training diverse model architectures; and (3) evaluating and interpreting model behavior. We designed EUGENe as a simple, flexible and extensible interface for streamlining and customizing end-to-end deep-learning sequence analyses, and illustrate these principles through application of the toolkit to three predictive modeling tasks. We hope that EUGENe represents a springboard towards a collaborative ecosystem for deep-learning applications in genomics research.

ORI-Explorer: a unified cell-specific tool for origin of replication sites prediction by feature fusion

MOTIVATION

The origins of replication sites (ORIs) are precise regions inside the DNA sequence where the replication process begins. These locations are critical for preserving the genome's integrity during cell division and guaranteeing the faithful transfer of genetic data from generation to generation. The advent of experimental techniques has aided in the discovery of ORIs in many species. Experimentation, on the other hand, is often more time-consuming and pricey than computational approaches, and it necessitates specific equipment and knowledge. Recently, ORI sites have been predicted using computational techniques like motif-based searches and artificial intelligence algorithms based on sequence characteristics and chromatin states.

RESULTS

In this article, we developed ORI-Explorer, a unique artificial intelligence-based technique that combines multiple feature engineering techniques to train CatBoost Classifier for recognizing ORIs from four distinct eukaryotic species. ORI-Explorer was created by utilizing a unique combination of three traditional feature-encoding techniques and a feature set obtained from a deep-learning neural network model. The ORI-Explorer has significantly outperformed current predictors on the testing dataset. Furthermore, by employing the sophisticated SHapley Additive exPlanation method, we give crucial insights that aid in comprehending model success, highlighting the most relevant features vital for forecasting cell-specific ORIs. ORI-Explorer is also intended to aid community-wide attempts in discovering potential ORIs and developing innovative verifiable biological hypotheses.

AVAILABILITY AND IMPLEMENTATION

The used datasets along with the source code are made available through https://github.com/Z-Abbas/ORI-Explorer and https://zenodo.org/record/8358679.

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.

CRMnet: A deep learning model for predicting gene expression from large regulatory sequence datasets

Recent large datasets measuring the gene expression of millions of possible gene promoter sequences provide a resource to design and train optimized deep neural network architectures to predict expression from sequences. High predictive performance due to the modeling of dependencies within and between regulatory sequences is an enabler for biological discoveries in gene regulation through model interpretation techniques. To understand the regulatory code that delineates gene expression, we have designed a novel deep-learning model (CRMnet) to predict gene expression in Saccharomyces cerevisiae. Our model outperforms the current benchmark models and achieves a Pearson correlation coefficient of 0.971 and a mean squared error of 3.200. Interpretation of informative genomic regions determined from model saliency maps, and overlapping the saliency maps with known yeast motifs, supports that our model can successfully locate the binding sites of transcription factors that actively modulate gene expression. We compare our model's training times on a large compute cluster with GPUs and Google TPUs to indicate practical training times on similar datasets.

Biological Sequence Classification: A Review on Data and General Methods

With the rapid development of biotechnology, the number of biological sequences has grown exponentially. The continuous expansion of biological sequence data promotes the application of machine learning in biological sequences to construct predictive models for mining biological sequence information. There are many branches of biological sequence classification research. In this review, we mainly focus on the function and modification classification of biological sequences based on machine learning. Sequence-based prediction and analysis are the basic tasks to understand the biological functions of DNA, RNA, proteins, and peptides. However, there are hundreds of classification models developed for biological sequences, and the quite varied specific methods seem dizzying at first glance. Here, we aim to establish a long-term support website (http://lab.malab.cn/~acy/BioseqData/home.html), which provides readers with detailed information on the classification method and download links to relevant datasets. We briefly introduce the steps to build an effective model framework for biological sequence data. In addition, a brief introduction to single-cell sequencing data analysis methods and applications in biology is also included. Finally, we discuss the current challenges and future perspectives of biological sequence classification research.

Identification of DNA-binding proteins via Multi-view LSSVM with independence criterion

DNA-binding proteins actively participate in life activities such as DNA replication, recombination, gene expression and regulation and play a prominent role in these processes. As DNA-binding proteins continue to be discovered and increase, it is imperative to design an efficient and accurate identification tool. Considering the time-consuming and expensive traditional experimental technology and the insufficient number of samples in the biological computing method based on structural information, we proposed a machine learning algorithm based on sequence information to identify DNA binding proteins, named multi-view Least Squares Support Vector Machine via Hilbert-Schmidt Independence Criterion (multi-view LSSVM via HSIC). This method took 6 feature sets as multi-view input and trains a single view through the LSSVM algorithm. Then, we integrated HSIC into LSSVM as a regular term to reduce the dependence between views and explored the complementary information of multiple views. Subsequently, we trained and coordinated the submodels and finally combined the submodels in the form of weights to obtain the final prediction model. On training set PDB1075, the prediction results of our model were better than those of most existing methods. Independent tests are conducted on the datasets PDB186 and PDB2272. The accuracy of the prediction results was 85.5% and 79.36%, respectively. This result exceeded the current state-of-the-art methods, which showed that the multi-view LSSVM via HSIC can be used as an efficient predictor.

DeepFusionDTA: Drug-Target Binding Affinity Prediction With Information Fusion and Hybrid Deep-Learning Ensemble Model

Identification of drug-target interaction (DTI) is the most important issue in the broad field of drug discovery. Using purely biological experiments to verify drug-target binding profiles takes lots of time and effort, so computational technologies for this task obviously have great benefits in reducing the drug search space. Most of computational methods to predict DTI are proposed to solve a binary classification problem, which ignore the influence of binding strength. Therefore, drug-target binding affinity prediction is still a challenging issue. Currently, lots of studies only extract sequence information that lacks feature-rich representation, but we consider more spatial features in order to merge various data in drug and target spaces. In this study, we propose a two-stage deep neural network ensemble model for detecting drug-target binding affinity, called DeepFusionDTA, via various information analysis modules. First stage is to utilize sequence and structure information to generate fusion feature map of candidate protein and drug pair through various analysis modules based deep learning. Second stage is to apply bagging-based ensemble learning strategy for regression prediction, and we obtain outstanding results by combining the advantages of various algorithms in efficient feature abstraction and regression calculation. Importantly, we evaluate our novel method, DeepFusionDTA, which delivers 1.5 percent CI increase on KIBA dataset and 1.0 percent increase on Davis dataset, by comparing with existing prediction tools, DeepDTA. Furthermore, the ideas we have offered can be applied to in-silico screening of the interaction space, to provide novel DTIs which can be experimentally pursued. The codes and data are available from https://github.com/guofei-tju/DeepFusionDTA.

scEpiLock: A Weakly Supervised Learning Framework for cis-Regulatory Element Localization and Variant Impact Quantification for Single-Cell Epigenetic Data

Recent advances in single-cell transposase-accessible chromatin using a sequencing assay (scATAC-seq) allow cellular heterogeneity dissection and regulatory landscape reconstruction with an unprecedented resolution. However, compared to bulk-sequencing, its ultra-high missingness remarkably reduces usable reads in each cell type, resulting in broader, fuzzier peak boundary definitions and limiting our ability to pinpoint functional regions and interpret variant impacts precisely. We propose a weakly supervised learning method, scEpiLock, to directly identify core functional regions from coarse peak labels and quantify variant impacts in a cell-type-specific manner. First, scEpiLock uses a multi-label classifier to predict chromatin accessibility via a deep convolutional neural network. Then, its weakly supervised object detection module further refines the peak boundary definition using gradient-weighted class activation mapping (Grad-CAM). Finally, scEpiLock provides cell-type-specific variant impacts within a given peak region. We applied scEpiLock to various scATAC-seq datasets and found that it achieves an area under receiver operating characteristic curve (AUC) of ~0.9 and an area under precision recall (AUPR) above 0.7. Besides, scEpiLock's object detection condenses coarse peaks to only ⅓ of their original size while still reporting higher conservation scores. In addition, we applied scEpiLock on brain scATAC-seq data and reported several genome-wide association studies (GWAS) variants disrupting regulatory elements around known risk genes for Alzheimer's disease, demonstrating its potential to provide cell-type-specific biological insights in disease studies.

Identifying and Classifying Enhancers by Dinucleotide-Based Auto-Cross Covariance and Attention-Based Bi-LSTM

Enhancers are a class of noncoding DNA elements located near structural genes. In recent years, their identification and classification have been the focus of research in the field of bioinformatics. However, due to their high free scattering and position variability, although the performance of the prediction model has been continuously improved, there is still a lot of room for progress. In this paper, density-based spatial clustering of applications with noise (DBSCAN) was used to screen the physicochemical properties of dinucleotides to extract dinucleotide-based auto-cross covariance (DACC) features; then, the features are reduced by feature selection Python toolkit MRMD 2.0. The reduced features are input into the random forest to identify enhancers. The enhancer classification model was built by word2vec and attention-based Bi-LSTM. Finally, the accuracies of our enhancer identification and classification models were 77.25% and 73.50%, respectively, and the Matthews' correlation coefficients (MCCs) were 0.5470 and 0.4881, respectively, which were better than the performance of most predictors.

Supervised promoter recognition: a benchmark framework

Motivation

Deep learning has become a prevalent method in identifying genomic regulatory sequences such as promoters. In a number of recent papers, the performance of deep learning models has continually been reported as an improvement over alternatives for sequence-based promoter recognition. However, the performance improvements in these models do not account for the different datasets that models are evaluated on. The lack of a consensus dataset and procedure for benchmarking purposes has made the comparison of each model’s true performance difficult to assess.

Results

We present a framework called Supervised Promoter Recognition Framework (‘SUPR REF’) capable of streamlining the complete process of training, validating, testing, and comparing promoter recognition models in a systematic manner. SUPR REF includes the creation of biologically relevant benchmark datasets to be used in the evaluation process of deep learning promoter recognition models. We showcase this framework by comparing the models’ performances on alternative datasets, and properly evaluate previously published models on new benchmark datasets. Our results show that the reliability of deep learning ab initio promoter recognition models on eukaryotic genomic sequences is still not at a sufficient level, as overall performance is still low. These results originate from a subset of promoters, the well-known RNA Polymerase II core promoters. Furthermore, given the observational nature of these data, cross-validation results from small promoter datasets need to be interpreted with caution.

The evolution, evolvability and engineering of gene regulatory DNA

Mutations in non-coding regulatory DNA sequences can alter gene expression, organismal phenotype and fitness^1-3. Constructing complete fitness landscapes, in which DNA sequences are mapped to fitness, is a long-standing goal in biology, but has remained elusive because it is challenging to generalize reliably to vast sequence spaces^4-6. Here we build sequence-to-expression models that capture fitness landscapes and use them to decipher principles of regulatory evolution. Using millions of randomly sampled promoter DNA sequences and their measured expression levels in the yeast Saccharomyces cerevisiae, we learn deep neural network models that generalize with excellent prediction performance, and enable sequence design for expression engineering. Using our models, we study expression divergence under genetic drift and strong-selection weak-mutation regimes to find that regulatory evolution is rapid and subject to diminishing returns epistasis; that conflicting expression objectives in different environments constrain expression adaptation; and that stabilizing selection on gene expression leads to the moderation of regulatory complexity. We present an approach for using such models to detect signatures of selection on expression from natural variation in regulatory sequences and use it to discover an instance of convergent regulatory evolution. We assess mutational robustness, finding that regulatory mutation effect sizes follow a power law, characterize regulatory evolvability, visualize promoter fitness landscapes, discover evolvability archetypes and illustrate the mutational robustness of natural regulatory sequence populations. Our work provides a general framework for designing regulatory sequences and addressing fundamental questions in regulatory evolution.

iPro-GAN: A novel model based on generative adversarial learning for identifying promoters and their strength

BACKGROUND AND OBJECTIVE

Promoter is a component of the gene, which can specifically bind with RNA polymerase and determine where transcription starts, and also determine the transcription efficiency of the gene. Promoters can be divided into strong promoters and weak promoters because their structures and the interaction time interval are quite different. The functional variation of the promoter can lead to a variety of diseases. Therefore, identifying promoters and their strength is necessary and has important biological significance. A novel and promising model based on deep learning is proposed to achieve it.

METHODS

In this work, we build a power model named iPro-GAN for identification of promoters and their strength. First, we collect benchmark datasets and independent datasets for training and testing. Then, Moran-based spatial auto-cross correlation method is used as feature extraction method. Finally, deep convolution generative adversarial network with 10-fold cross validation is applied for classifying. The first layer of the model is used to identify the promoter and the second layer is used to determine its type.

RESULTS

On the benchmark data set, the accuracy of the first layer predictor is 93.15%, and the accuracy of the second layer predictor is 92.30%. On the independent data set, the accuracy of the first layer predictor is 86.77%, and the accuracy of the second layer predictor is 91.66%. In particular, breakthrough progress has been made in the identification of promoters' strength.

CONCLUSIONS

These results are far higher than the existing best predictor, which indicate that our model is serviceable and practicable to identify promoters and their strength. Furthermore, the datasets and source codes are available from this link: https://github.com/Bovbene/iPro-GAN.

The Characterization of Structure and Prediction for Aquaporin in Tumour Progression by Machine Learning

Recurrence and new cases of cancer constitute a challenging human health problem. Aquaporins (AQPs) can be expressed in many types of tumours, including the brain, breast, pancreas, colon, skin, ovaries, and lungs, and the histological grade of cancer is positively correlated with AQP expression. Therefore, the identification of aquaporins is an area to explore. Computational tools play an important role in aquaporin identification. In this research, we propose reliable, accurate and automated sequence predictor iAQPs-RF to identify AQPs. In this study, the feature extraction method was 188D (global protein sequence descriptor, GPSD). Six common classifiers, including random forest (RF), NaiveBayes (NB), support vector machine (SVM), XGBoost, logistic regression (LR) and decision tree (DT), were used for AQP classification. The classification results show that the random forest (RF) algorithm is the most suitable machine learning algorithm, and the accuracy was 97.689%. Analysis of Variance (ANOVA) was used to analyse these characteristics. Feature rank based on the ANOVA method and IFS strategy was applied to search for the optimal features. The classification results suggest that the 26th feature (neutral/hydrophobic) and 21st feature (hydrophobic) are the two most powerful and informative features that distinguish AQPs from non-AQPs. Previous studies reported that plasma membrane proteins have hydrophobic characteristics. Aquaporin subcellular localization prediction showed that all aquaporins were plasma membrane proteins with highly conserved transmembrane structures. In addition, the 3D structure of aquaporins was consistent with the localization results. Therefore, these studies confirmed that aquaporins possess hydrophobic properties. Although aquaporins are highly conserved transmembrane structures, the phylogenetic tree shows the diversity of aquaporins during evolution. The PCA showed that positive and negative samples were well separated by 54D features, indicating that the 54D feature can effectively classify aquaporins. The online prediction server is accessible at http://lab.malab.cn/∼acy/iAQP.

SNAREs-SAP: SNARE Proteins Identification With PSSM Profiles

Soluble N-ethylmaleimide sensitive factor activating protein receptor (SNARE) proteins are a large family of transmembrane proteins located in organelles and vesicles. The important roles of SNARE proteins include initiating the vesicle fusion process and activating and fusing proteins as they undergo exocytosis activity, and SNARE proteins are also vital for the transport regulation of membrane proteins and non-regulatory vesicles. Therefore, there is great significance in establishing a method to efficiently identify SNARE proteins. However, the identification accuracy of the existing methods such as SNARE CNN is not satisfied. In our study, we developed a method based on a support vector machine (SVM) that can effectively recognize SNARE proteins. We used the position-specific scoring matrix (PSSM) method to extract features of SNARE protein sequences, used the support vector machine recursive elimination correlation bias reduction (SVM-RFE-CBR) algorithm to rank the importance of features, and then screened out the optimal subset of feature data based on the sorted results. We input the feature data into the model when building the model, used 10-fold crossing validation for training, and tested model performance by using an independent dataset. In independent tests, the ability of our method to identify SNARE proteins achieved a sensitivity of 68%, specificity of 94%, accuracy of 92%, area under the curve (AUC) of 84%, and Matthew’s correlation coefficient (MCC) of 0.48. The results of the experiment show that the common evaluation indicators of our method are excellent, indicating that our method performs better than other existing classification methods in identifying SNARE proteins.

Identification of plant vacuole proteins by exploiting deep representation learning features

Plant vacuoles are the most important organelles for plant growth, development, and defense, and they play an important role in many types of stress responses. An important function of vacuole proteins is the transport of various classes of amino acids, ions, sugars, and other molecules. Accurate identification of vacuole proteins is crucial for revealing their biological functions. Several automatic and rapid computational tools have been proposed for the subcellular localization of proteins. Regrettably, they are not specific for the identification of plant vacuole proteins. To the best of our knowledge, there is only one computational software specifically trained for plant vacuolar proteins. Although its accuracy is acceptable, the prediction performance and stability of this method in practical applications can still be improved. Hence, in this study, a new predictor named iPVP-DRLF was developed to identify plant vacuole proteins specifically and effectively. This prediction software is designed using the light gradient boosting machine (LGBM) algorithm and hybrid features composed of classic sequence features and deep representation learning features. iPVP-DRLF achieved fivefold cross-validation and independent test accuracy values of 88.25 % and 87.16 %, respectively, both outperforming previous state-of-the-art predictors. Moreover, the blind dataset test results also showed that the performance of iPVP-DRLF was significantly better than the existing tools. The results of comparative experiments confirmed that deep representation learning features have an advantage over other classic sequence features in the identification of plant vacuole proteins. We believe that iPVP-DRLF would serve as an effective computational technique for plant vacuole protein prediction and facilitate related future research. The online server is freely accessible at https://lab.malab.cn/~acy/iPVP-DRLF. In addition, the source code and datasets are also accessible at https://github.com/jiaoshihu/iPVP-DRLF.

DeepAc4C: a convolutional neural network model with hybrid features composed of physicochemical patterns and distributed representation information for identification of N4-acetylcytidine in mRNA

MOTIVATION

N4-acetylcytidine (ac4C) is the only acetylation modification that has been characterized in eukaryotic RNA, and is correlated with various human diseases. Laboratory identification of ac4C is complicated by factors, such as sample hydrolysis and high cost. Unfortunately, existing computational methods to identify ac4C do not achieve satisfactory performance.

RESULTS

We developed a novel tool, DeepAc4C, which identifies ac4C using convolutional neural networks (CNNs) using hybrid features composed of physicochemical patterns and a distributed representation of nucleic acids. Our results show that the proposed model achieved better and more balanced performance than existing predictors. Furthermore, we evaluated the effect that specific features had on the model predictions and their interaction effects. Several interesting sequence motifs specific to ac4C were identified.

AVAILABILITY AND IMPLEMENTATION

The webserver is freely accessible at https://ac4c.webmalab.cn/, the source code and datasets are accessible at Zenodo with URL https://doi.org/10.5281/zenodo.5138047 and Github with URL https://github.com/wangchao-malab/DeepAc4C.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Pseudo-188D: Phage Protein Prediction Based on a Model of Pseudo-188D

Phages have seriously affected the biochemical systems of the world, and not only are phages related to our health, but medical treatments for many cancers and skin infections are related to phages; therefore, this paper sought to identify phage proteins. In this paper, a Pseudo-188D model was established. The digital features of the phage were extracted by PseudoKNC, an appropriate vector was selected by the AdaBoost tool, and features were extracted by 188D. Then, the extracted digital features were combined together, and finally, the viral proteins of the phage were predicted by a stochastic gradient descent algorithm. Our model effect reached 93.4853%. To verify the stability of our model, we randomly selected 80% of the downloaded data to train the model and used the remaining 20% of the data to verify the robustness of our model.

iAIPs: Identifying Anti-Inflammatory Peptides Using Random Forest

Recently, several anti-inflammatory peptides (AIPs) have been found in the process of the inflammatory response, and these peptides have been used to treat some inflammatory and autoimmune diseases. Therefore, identifying AIPs accurately from a given amino acid sequences is critical for the discovery of novel and efficient anti-inflammatory peptide-based therapeutics and the acceleration of their application in therapy. In this paper, a random forest-based model called iAIPs for identifying AIPs is proposed. First, the original samples were encoded with three feature extraction methods, including g-gap dipeptide composition (GDC), dipeptide deviation from the expected mean (DDE), and amino acid composition (AAC). Second, the optimal feature subset is generated by a two-step feature selection method, in which the feature is ranked by the analysis of variance (ANOVA) method, and the optimal feature subset is generated by the incremental feature selection strategy. Finally, the optimal feature subset is inputted into the random forest classifier, and the identification model is constructed. Experiment results showed that iAIPs achieved an AUC value of 0.822 on an independent test dataset, which indicated that our proposed model has better performance than the existing methods. Furthermore, the extraction of features for peptide sequences provides the basis for evolutionary analysis. The study of peptide identification is helpful to understand the diversity of species and analyze the evolutionary history of species.

KK-DBP: A Multi-Feature Fusion Method for DNA-Binding Protein Identification Based on Random Forest

DNA-binding protein (DBP) is a protein with a special DNA binding domain that is associated with many important molecular biological mechanisms. Rapid development of computational methods has made it possible to predict DBP on a large scale; however, existing methods do not fully integrate DBP-related features, resulting in rough prediction results. In this article, we develop a DNA-binding protein identification method called KK-DBP. To improve prediction accuracy, we propose a feature extraction method that fuses multiple PSSM features. The experimental results show a prediction accuracy on the independent test dataset PDB186 of 81.22%, which is the highest of all existing methods.

iTTCA-RF: a random forest predictor for tumor T cell antigens

BACKGROUND

Cancer is one of the most serious diseases threatening human health. Cancer immunotherapy represents the most promising treatment strategy due to its high efficacy and selectivity and lower side effects compared with traditional treatment. The identification of tumor T cell antigens is one of the most important tasks for antitumor vaccines development and molecular function investigation. Although several machine learning predictors have been developed to identify tumor T cell antigen, more accurate tumor T cell antigen identification by existing methodology is still challenging.

METHODS

In this study, we used a non-redundant dataset of 592 tumor T cell antigens (positive samples) and 393 tumor T cell antigens (negative samples). Four types feature encoding methods have been studied to build an efficient predictor, including amino acid composition, global protein sequence descriptors and grouped amino acid and peptide composition. To improve the feature representation ability of the hybrid features, we further employed a two-step feature selection technique to search for the optimal feature subset. The final prediction model was constructed using random forest algorithm.

RESULTS

Finally, the top 263 informative features were selected to train the random forest classifier for detecting tumor T cell antigen peptides. iTTCA-RF provides satisfactory performance, with balanced accuracy, specificity and sensitivity values of 83.71%, 78.73% and 88.69% over tenfold cross-validation as well as 73.14%, 62.67% and 83.61% over independent tests, respectively. The online prediction server was freely accessible at http://lab.malab.cn/~acy/iTTCA .

CONCLUSIONS

We have proven that the proposed predictor iTTCA-RF is superior to the other latest models, and will hopefully become an effective and useful tool for identifying tumor T cell antigens presented in the context of major histocompatibility complex class I.

PHIAF: prediction of phage-host interactions with GAN-based data augmentation and sequence-based feature fusion

Phage therapy has become one of the most promising alternatives to antibiotics in the treatment of bacterial diseases, and identifying phage-host interactions (PHIs) helps to understand the possible mechanism through which a phage infects bacteria to guide the development of phage therapy. Compared with wet experiments, computational methods of identifying PHIs can reduce costs and save time and are more effective and economic. In this paper, we propose a PHI prediction method with a generative adversarial network (GAN)-based data augmentation and sequence-based feature fusion (PHIAF). First, PHIAF applies a GAN-based data augmentation module, which generates pseudo PHIs to alleviate the data scarcity. Second, PHIAF fuses the features originated from DNA and protein sequences for better performance. Third, PHIAF utilizes an attention mechanism to consider different contributions of DNA/protein sequence-derived features, which also provides interpretability of the prediction model. In computational experiments, PHIAF outperforms other state-of-the-art PHI prediction methods when evaluated via 5-fold cross-validation (AUC and AUPR are 0.88 and 0.86, respectively). An ablation study shows that data augmentation, feature fusion and an attention mechanism are all beneficial to improve the prediction performance of PHIAF. Additionally, four new PHIs with the highest PHIAF score in the case study were verified by recent literature. In conclusion, PHIAF is a promising tool to accelerate the exploration of phage therapy.

Prediction of presynaptic and postsynaptic neurotoxins based on feature extraction

A neurotoxin is essentially a protein that mainly acts on the nervous system; it has a selective toxic effect on the central nervous system and neuromuscular nodes, can cause muscle paralysis and respiratory paralysis, and has strong lethality. According to their principle of action, neurotoxins are divided into presynaptic neurotoxins and postsynaptic neurotoxins. Correctly identifying presynaptic and postsynaptic nerve toxins provides important clues for future drug development and the discovery of drug targets. Therefore, a predictive model, Neu_LR, was constructed in this paper. The monoMonokGap method was used to extract the frequency characteristics of presynaptic and postsynaptic neurotoxin sequences and carry out feature selection, then, based on the important features obtained after dimensionality reduction, the prediction model Neu_LR was constructed using a logistic regression algorithm, and ten-fold cross-validation and independent test set validation were used. The final accuracy rates were 99.6078 and 94.1176%, respectively, which proved that the Neu_LR model had good predictive performance and robustness, and could meet the prediction requirements of presynaptic and postsynaptic neurotoxins. The data and source code of the model can be freely download from https://github.com/gyx123681/.

Application of Machine Learning for Drug-Target Interaction Prediction

Exploring drug–target interactions by biomedical experiments requires a lot of human, financial, and material resources. To save time and cost to meet the needs of the present generation, machine learning methods have been introduced into the prediction of drug–target interactions. The large amount of available drug and target data in existing databases, the evolving and innovative computer technologies, and the inherent characteristics of various types of machine learning have made machine learning techniques the mainstream method for drug–target interaction prediction research. In this review, details of the specific applications of machine learning in drug–target interaction prediction are summarized, the characteristics of each algorithm are analyzed, and the issues that need to be further addressed and explored for future research are discussed. The aim of this review is to provide a sound basis for the construction of high-performance models.

i6mA-VC: A Multi-Classifier Voting Method for the Computational Identification of DNA N6-methyladenine Sites

DNA N6-methyladenine (6 mA), as an essential component of epigenetic modification, cannot be neglected in genetic regulation mechanism. The efficient and accurate prediction of 6 mA sites is beneficial to the development of biological genetics. Biochemical experimental methods are considered to be time-consuming and laborious. Most of the established machine learning methods have a single dataset. Although some of them have achieved cross-species prediction, their results are not satisfactory. Therefore, we designed a novel statistical model called i6mA-VC to improve the accuracy for 6 mA sites. On the one hand, kmer and binary encoding are applied to extract features, and then gradient boosting decision tree (GBDT) embedded method is applied as the feature selection strategy. On the other hand, DNA sequences are represented by vectors through the feature extraction method of ring-function-hydrogen-chemical properties (RFHCP) and the feature selection strategy of ExtraTree. After fusing the two optimal features, a voting classifier based on gradient boosting decision tree (GBDT), light gradient boosting machine (LightGBM) and multilayer perceptron classifier (MLPC) is constructed for final classification and prediction. The accuracy of Rice dataset and M.musculus dataset with five-fold cross-validation are 0.888 and 0.967, respectively. The cross-species dataset is selected as independent testing dataset, and the accuracy reaches 0.848. Through rigorous experiments, it is demonstrated that the proposed predictor is convincing and applicable. The development of i6mA-VC predictor will become an effective way for the recognition of N6-methyladenine sites, and it will also be beneficial for biological geneticists to further study gene expression and DNA modification. In addition, an accessible web-server for i6mA-VC is available from http://www.zhanglab.site/ .

Identification of long noncoding RNAs with machine learning methods: a review

Long noncoding RNAs (lncRNAs) are noncoding RNAs with a length greater than 200 nucleotides. Studies have shown that they play an important role in many life activities. Dozens of lncRNAs have been characterized to some extent, and they are reported to be related to the development of diseases in a variety of cells. However, the biological functions of most lncRNAs are currently still unclear. Therefore, accurately identifying and predicting lncRNAs would be helpful for research on their biological functions. Due to the disadvantages of high cost and high resource-intensiveness of experimental methods, scientists have developed numerous computational methods to identify and predict lncRNAs in recent years. In this paper, we systematically summarize the machine learning-based lncRNAs prediction tools from several perspectives, and discuss the challenges and prospects for the future work.