RNA plays a pivotal role in diverse cellular processes, and the rational design of functional RNA sequences is central to advancing RNA engineering. While deep generative models have shown significant promise, their generated sequences frequently lack the structural accuracy and evolutionary fidelity required for biological functionality. To address this challenge, we introduced VAERNAGen, a novel variational autoencoder-based framework for de novo generation of RNA family sequences. VAERNAGen's core innovation is a joint representation that encode aligned nucleotide sequences and their secondary structures into an 11-channel L × L two-dimensional matrix. This image-like format enables 2D convolutional neural networks to effectively learn the spatial interplay between sequence conservation and structural variation. Evaluated on two canonical Rfam families-RF00001 (5S ribosomal RNA) and RF00005 (tRNA), VAERNAGen outperformed the current state-of-the-art grammar-based method, achieving significantly higher median bit scores (105.25 vs. 89.34 for RF00001; 58.24 vs. 50.02 for RF00005). Generated sequences also exhibited greater nucleotide-level similarity to natural seed alignments and lower variance, reflecting enhanced biological plausibility and reproducibility. Together, these results establish VAERNAGen as a state-of-the-art method for de novo generation of RNA family sequences.

Multi-view clustering aims to improve clustering performance by jointly considering all available views and utilizing the complementary and consistent information of multiple heterogeneous views. However, multi-view data often have problems such as noise, missing points, and poor adaptability to dynamic data. Existing methods face challenges such as view heterogeneity and the inability to model nonlinear inter-view relationships. To address these issues, a dual-branch Bayesian adversarial learning for multi-view clustering (DBAMVC) model is proposed for aligning nonlinear and heterogeneous views, making up for the deficiency of traditional canonical correlation analysis (CCA) in modeling nonlinear relationships. This model introduces adversarial learning in the multi-view fusion branch, distinguishing the fusion results of actual data and generated data through a discriminator, and guiding the latent representation module in reverse to optimize the feature representation. By introducing the collaborative mechanism of adversarial learning and Bayesian optimization, a dual-branch collaborative optimization framework is constructed. Adversarial learning provides gradient signals for fusion goals. Bayesian inference constrains the direction of parameter update from a probabilistic perspective, and both collaborate in the optimization to solve the problem of separating representation learning from the fusion goal in traditional multi-view clustering while enhancing adaptability to dynamic data. Experimental results show that the method proposed achieves superior performance compared to the comparison methods in the evaluation indexes used.

Spectral clustering is a basic method in unsupervised learning based on graphs. This method can extract the inherent cluster structure from the characteristic space of the graph Laplacian. However, different spectral components have different effects on cluster quality. In robust multi-view clustering, the role of these components has not been fully understood. This article first gives a theoretical analysis. Analysis shows that the low-frequency part retains the main cluster subspace. High-frequency truncation can reduce the sensitivity of the model to disturbances without destroying the spectral structure of the core. Then, we studied the frequency characteristics of spectral clustering through experiments. Experiments reveal an obvious phenomenon: removing low-frequency components will seriously weaken the clustering effect, while appropriately suppressing a certain amount of high-frequency components will usually improve robustness. Inspired by these findings, we propose a new framework with theoretical support, namely Multi-View Spectral Clustering via High-Frequency Truncation and Graph Wavelet Modulation(MST-WM). The specific practice of the framework is that the graph Laplacian corresponding to each view is first truncated to remove the unstable high-frequency component. Then, the obtained Spectral embedding is fused into a consistent representation. The representation is further refined by graph wavelet modulation, which can not only enhance the smooth structural information, but also suppress residual noise. A large number of experimental results on 10 benchmark data sets show that the proposed method is stable and superior to a variety of competitive baselines in terms of ACC, NMI and purity. These results show that the unequal effect of explicit modeling spectral components provides a principled and effective path for achieving robust multi-view clustering.

Introduction Emerging evidence highlights the critical role of gut microbes in the pathogenesis of autism spectrum disorder through interactions within the microbiota-gut-brain axis.Methods This study conducted systematic bioinformatics analysis on 16S rRNA (V3-V4 region) sequencing data from 78 children with autism spectrum disorder and 68 neurotypical children from the NCBI database.Results Microbial composition analysis revealed significant changes between patients with autism spectrum disorder and healthy controls across phylum-level and genus-level classifications. Elevated Bacteroidota/Firmicutes ratio was found in children with autism spectrum disorder compared to healthy controls. Microbial diversity analysis showed significant intergroup divergence: autism spectrum disorder subjects exhibited significantly elevated alpha-diversity, mainly reflected in the Chao1 index. Moreover, significant structural separation was observed between these two groups. LefSe analysis demonstrated distinct enrichment of Bacteroidota and Actinobacteriota at the phylum level, and Bacteroides, Prevotella, and Megamonas at the genus level in children with autism spectrum disorder. Conversely, control-associated enrichment of Firmicutes (phylum), along with genera Ruminococcus and Escherichia-Shigella, showed depletion in children with autism spectrum disorder. Functional prediction via PICRUSt2 demonstrated autism spectrum disorder-specific upregulation of cofactor and vitamin metabolism, contrasting with enrichment of terpenoid, polyketide, and xenobiotic degradation pathways in healthy controls.Discussion These findings substantiate the microbiota-gut-brain axis hypothesis in the pathogenesis of autism spectrum disorder, revealing microbial composition and functional shifts potentially linked to nutritional compensation and inflammatory modulation.Conclusion Gut microbial dysbiosis in autism spectrum disorder, characterized by altered microbial diversity and community structure, impairs xenobiotic metabolism, leading to neurotoxin accumulation that exacerbates core symptoms.

Temporal networks model the evolution of complex systems, and tracking dynamic communities within them is critical for understanding the underlying systems, where both connectivity (clustering accuracy) and temporality (clustering drift) must be addressed simultaneously. Current algorithms cannot precisely model the temporality of networks at the feature level for each vertex, thereby failing to balance clustering accuracy and drift. To attack this limitation, a novel Feature Separation-based Evolution Nonnegative Matrix Factorization algorithm (Fs-ENMF) for identifying evolving communities of temporal networks is proposed. Fs-ENMF simultaneously learns feature for the conserved and dynamic structure of networks. Specifically, Fs-ENMF separates the static and dynamic parts of vertex features, where relations between the shared and conserved structure are explicitly exploited. Furthermore, dynamics of networks are measured at the feature level for each vertex, which is formulated as /2,1-norm constraint. Moreover, temporality is exploited by integrating features at the previous and current time. Finally, Fs-ENMF integrates the feature decomposition, dynamics of vertices, and temporality procedures with an optimization function and rules. Experiments demonstrate the superiority of Fs-ENMF in terms of accuracy, indicating that discriminating the static and dynamic features is promising for tracking communities in temporal networks.

Cancer is a serious and complex disease caused by uncontrolled cell growth and is becoming one of the leading causes of death worldwide. Anticancer peptides (ACPs), as a bioactive peptide with lower toxicity, emerge as a promising means of effectively treating cancer. Identifying ACPs is challenging due to the limitation of experimental conditions. To address this, we proposed a dual-channel-based deep learning method, termed ACP-DPE, for ACP prediction. The ACP-DPE consisted of two parallel channels: one was an embedding layer followed by the bi-directional gated recurrent unit (Bi-GRU) module, and the other was an adaptive embedding layer followed by the dilated convolution module. The Bi-GRU module captured the peptide sequence dependencies, whereas the dilated convolution module characterised the local relationship of amino acids. Experimental results show that ACP-DPE achieves an accuracy of 82.81% and a sensitivity of 86.63%, surpassing the state-of-the-art method by 3.86% and 5.1%, respectively. These findings demonstrate the effectiveness of ACP-DPE for ACP prediction and highlight its potential as a valuable tool in cancer treatment research.

Multi-view clustering garners extensive attention due to the increasing significance of accumulated multi-view data. However, current algorithms model diversity and consistency of various views with sparsity constraints, which fail to fully explore structure and relations among multiple views. To address this limitation, we propose a novel One-step Multi-view Spectral Clustering with Subspaces Fusion on Grassmann Manifold (OMSC_SFGM), where similarity graph construction, consistency and diversity learning, and clustering are incorporated into a unified joint learning model. Specifically, OMSC_SFGM constructs a similarity graph for each view using self-representation learning, which is decomposed into diverse and consistent parts. To precisely model the diversity of views, OMSC_SFGM characterizes the diversity of each view using the subspace of other views, where subspace distance on the Grassmann manifold is adopted to quantify correlations among views. This explicitly addresses the relationships between views, enhancing feature quality by maintaining geometric structures across views. Moreover, OMSC_SFGM proposes a one-step spectral strategy that integrates representation learning and clustering with spectral rotation. Comprehensive experiments on eleven real-world datasets demonstrate that the proposed algorithm outperforms baseline methods in multi-view clustering.

ABSTRACT Long non‐coding RNAs (lncRNAs) are closely associated with the regulation of gene expression, whose promoters play a crucial role in comprehensively understanding lncRNA regulatory mechanisms, functions and their roles in diseases. Due to limitations of the current techniques, accurately identifying lncRNA promoters remains a challenge. To address this challenge, we propose a support vector machine (SVM)–based method for predicting lncRNA promoters, called SVM‐LncRNAPro. This method uses position‐specific trinucleotide propensity based on single‐strand (PSTNPss) to encode the DNA sequences and employs an SVM as the learning algorithm. The SVM‐LncRNAPro achieves state‐of‐the‐art performance with reduced complexity. Additionally, experiments demonstrate that this method exhibits a strong generalisation ability. For the convenience of academic research, we have made the source code of SVM‐LncRNAPro publicly available. Researchers can download the code and perform the prediction of the lncRNA promoter via the following link: https://github.com/TG0F7/Prom/tree/master .

Long non-coding RNAs (lncRNAs) play key roles in numerous biological processes and are associated with various human diseases. High-throughput RNA sequencing (HTlncRNAs) has identified tens of thousands of lncRNAs across species, but only a small fraction have been functionally characterized. While the experimental validation of lncRNAs (EVlncRNAs) using low-throughput methods is increasing, the expensive costs limit the validation to a small subset of HTlncRNAs. Therefore, developing predictive tools to prioritize potentially functional lncRNAs for low-throughput validation is crucial. To address this need, we proposed EVlncRNA-net, a novel deep learning framework based on sequence language processing. This framework incorporates two representation learning modules: EVlncRNA-net (GCN) and EVlncRNA-net (CNN). EVlncRNA-net (GCN) introduces a novel graph construction method and a specialized node encoding technique. This module transforms lncRNA sequences into graphical formats and processes them using graph convolution. EVlncRNA-net (CNN) extracts features from one-hot encoded sequences via convolutional neural networks. Both modules ensure robust feature representation of lncRNA sequences. Tailored for humans, mice, and plants, EVlncRNA-net achieves prediction accuracies of 85.8 %, 83.1 %, and 85.4 %, respectively, outperforming existing methods. The platform is available at https://github.com/rice1ee/EVlncRNA_net/tree/master, serving as a valuable tool for prioritizing lncRNAs for experimental validation.

Identification of potential drug-target interactions (DTIs) is a crucial step in drug discovery and repurposing. Although deep learning effectively deciphers DTIs, most deep learning-based methods represent drug features from only a single perspective. Moreover, the fusion method of drug and protein features needs further refinement. To address the above two problems, in this study, we develop a novel end-to-end framework named DO-GMA for potential DTI identification by incorporating Depthwise Overparameterized convolutional neural network and the Gated Multihead Attention mechanism with shared-learned queries and bilinear model concatenation. DO-GMA first designs a depthwise overparameterized convolutional neural network to learn drug representations from their SMILES strings and protein representations from their amino acid sequences. Next, it extracts drug representations from their 2D molecular graphs through a graph convolutional network. Subsequently, it fuses drug and protein features by combining the gated attention mechanism and the multihead attention mechanism with shared-learned queries and bilinear model concatenation. Finally, it takes the fused drug-target features as inputs and builds a multilayer perceptron to classify unlabeled drug-target pairs (DTPs). DO-GMA was benchmarked against six newest DTI prediction methods (CPI-GNN, BACPI, CPGL, DrugBAN, BINDTI, and FOTF-CPI) under four different experimental settings on four DTI data sets (i.e., DrugBank, BioSNAP, C.elegans, and BindingDB). The results show that DO-GMA significantly outperformed the above six methods based on AUC, AUPR, accuracy, F1-score, and MCC. An ablation study, robust statistical analysis, sensitivity analysis of parameters, visualization of the fused features, computational cost analysis, and case analysis further validated the powerful DTI identification performance of DO-GMA. In addition, DO-GMA predicted that two drug-protein pairs (i.e., DB00568 and P06276, and DB09118 and Q9UQD0) could be interacting. DO-GMA is freely available at https://github.com/plhhnu/DO-GMA.

RNA N4-acetylcytidine (ac4C) is a crucial chemical modification involved in various biological processes, influencing RNA properties and functions. Accurate prediction of RNA ac4C sites is essential for understanding the roles of RNA molecules in gene expression and cellular regulation. While existing methods have made progress in ac4C site prediction, they still struggle with limited accuracy and generalization. To address these challenges, we propose DeepRNAac4C, a deep learning framework for RNA ac4C sites prediction. DeepRNAac4C integrates residual neural networks, convolutional neural networks (CNN), bidirectional long short-term memory networks (BiLSTM), and bidirectional gated recurrent units (BiGRU) to effectively capture both local and global sequence features. We extensively evaluated DeepRNAac4C against state-of-the-art methods using 10-fold cross-validation and independent tests. The results show that DeepRNAac4C outperforms existing approaches, achieving an accuracy of 0.8410. The proposed DeepRNAac4C improves predictive accuracy and model robustness, providing an effective tool for identifying RNA ac4C sites and deepening our understanding of RNA modifications and their functional roles in biological systems.

Phosphorylation, a reversible and widespread post-translational modification of proteins, is essential for numerous cellular processes. However, due to technical limitations, large-scale detection of phosphorylation sites, especially those infected by SARS-CoV-2, remains a challenging task. To address this gap, we propose a method called GBMPhos, a novel method that combines convolutional neural networks (CNNs) for extracting local features, gating mechanisms to selectively focus on relevant information, and a bi-directional gated recurrent unit (Bi-GRU) to capture long-range dependencies within protein sequences. GBMPhos leverages a comprehensive set of features, including sequence encoding, physicochemical properties, and structural information, to provide an in-depth analysis of phosphorylation sites. We conducted an extensive comparison of GBMPhos with traditional machine learning algorithms and state-of-the-art methods. Experimental results demonstrate the superiority of GBMPhos over existing methods. The visualization analysis further highlights its effectiveness and efficiency. Additionally, we have established a free web server platform to help researchers explore phosphorylation in SARS-CoV-2 infections. The source code of GBMPhos is publicly available on GitHub.

The significance of low-quality data in unlabeled medical images is always underestimated. We believe that these underestimated data contain valuable information that remains largely unexplored. We present a novel uncertainty-guided different levels of pseudolabels (UDLP) framework to explore the underestimated data in medical images. The framework consists of a student-teacher model that uses uncertainty to classify the pseudolabels predicted by the teacher model into three levels: high confidence, low confidence, and unreliability. The student model learns directly from high-confidence pseudolabels. By using the confident learning method in low-confidence pseudolabels, the teacher model corrects the noisy labels in low-confidence voxels to provide positive feature information for the student model. We design a method for removing unreliable pseudolabels, to further enhance model's generalizability. The proposed framework UDLP is evaluated on two datasets and demonstrates superior performance compared to other state-of-the-art methods.

N6-methyladenosine (m6A) is the most prevalent, abundant, and conserved internal modification in the eukaryotic messenger RNA (mRNAs) and plays a crucial role in the cellular process. Although more than ten methods were developed for m6A detection over the past decades, there were rooms left to improve the predictive accuracy and the efficiency. In this paper, we proposed an improved method for predicting m6A modification sites, which was based on bi-directional gated recurrent unit (Bi-GRU) and convolutional neural networks (CNN), called Deepm6A-MT. The Deepm6A-MT has two input channels. One is to use an embedding layer followed by the Bi-GRU and then by the CNN, and another is to use one-hot encoding, dinucleotide one-hot encoding, and nucleotide chemical property codes. We trained and evaluated the Deepm6A-MT both by the 5-fold cross-validation and the independent test. The empirical tests showed that the Deepm6A-MT achieved the state of the art performance. In addition, we also conducted the cross-species and the cross-tissues tests to further verify the Deepm6A-MT for effectiveness and efficiency. Finally, for the convenience of academic research, we deployed the Deepm6A-MT to the web server, which is accessed at the URL http://www.biolscience.cn/Deepm6A-MT/.

Clustering dynamic networks has gained popularity due to the need to analyze complex systems that evolve over time, which cannot be fully characterized by traditional static models. It is highly non-trivial in comparison to clustering static network since it requires simultaneously to balance clustering accuracy and clustering drift, where clustering accuracy measures how clustering reflects structure of graph at current time, and clustering drift quantifies how clustering smoothes historical snapshot(s). In this study, we propose an algorithm clustering dynamic network by discriminating roles of vertices and capturing temporality with subsequent feature projection (CDN-DRCT). Specifically, clustering accuracy is achieved by factorizing high-order matrix of slice at current time, and vertices are divided into static and dynamic ones by the reconstruction errors. Finally, the proposed algorithm measures temporality of networks with a projection matrix, which connects subsequent features at the previous and current time, thereby enhancing clustering drift of clusters. In this case, temporality of dynamic networks is characterized from vertex and global level, providing a better way to balance clustering accuracy and clustering drift. Experimental results on 10 typical dynamic networks demonstrate the proposed algorithm is superior to baselines in terms of accuracy as well efficiency.

DNA N6-adenine methylation (N6-methyladenine, 6mA) plays a key regulating role in the cellular processes. Precisely recognizing 6mA sites is of importance to further explore its biological functions. Although there are many developed computational methods for 6mA site prediction over the past decades, there is a large root left to improve. We presented a cross validation-based stacking ensemble model for 6mA site prediction, called 6mA-StackingCV. The 6mA-StackingCV is a type of meta-learning algorithm, which uses output of cross validation as input to the final classifier. The 6mA-StackingCV reached the state of the art performances in the Rosaceae independent test. Extensive tests demonstrated the stability and the flexibility of the 6mA-StackingCV. We implemented the 6mA-StackingCV as a user-friendly web application, which allows one to restrictively choose representations or learning algorithms. This application is freely available at http://www.biolscience.cn/6mA-stackingCV/ . The source code and experimental data is available at https://github.com/Xiaohong-source/6mA-stackingCV .

N4-methylcytosine (4mC) is an important epigenetic mechanism, which regulates many cellular processes such as cell differentiation and gene expression. The knowledge about the 4mC sites is a key foundation to exploring its roles. Due to the limitation of techniques, precise detection of 4mC is still a challenging task. In this paper, we presented a multi-scale convolution neural network (CNN) and adaptive embedding-based computational method for predicting 4mC sites in mouse genome, which was referred to as MultiScale-CNN-4mCPred. The MultiScale-CNN-4mCPred used adaptive embedding to encode nucleotides, and then utilized multi-scale CNNs as well as long short-term memory to extract more in-depth local properties and contextual semantics in the sequences. The MultiScale-CNN-4mCPred is an end-to-end learning method, which requires no sophisticated feature design. The MultiScale-CNN-4mCPred reached an accuracy of 81.66% in the 10-fold cross-validation, and an accuracy of 84.69% in the independent test, outperforming state-of-the-art methods. We implemented the proposed method into a user-friendly web application which is freely available at: http://www.biolscience.cn/MultiScale-CNN-4mCPred/ .

The epitope is the site where antigens and antibodies interact and is vital to understanding the immune system. Experimental identification of linear B-cell epitopes (BCEs) is expensive, is labor-consuming, and has a low throughput. Although a few computational methods have been proposed to address this challenge, there is still a long way to go for practical applications. We proposed a deep learning method called DeepLBCEPred for predicting linear BCEs, which consists of bi-directional long short-term memory (Bi-LSTM), feed-forward attention, and multi-scale convolutional neural networks (CNNs). We extensively tested the performance of DeepLBCEPred through cross-validation and independent tests on training and two testing datasets. The empirical results showed that the DeepLBCEPred obtained state-of-the-art performance. We also investigated the contribution of different deep learning elements to recognize linear BCEs. In addition, we have developed a user-friendly web application for linear BCEs prediction, which is freely available for all scientific researchers at: http://www.biolscience.cn/DeepLBCEPred/ .

Human leukocyte antigen (HLA) is closely involved in regulating the human immune system. Despite great advance in detecting classical HLA Class I binders, there are few methods or toolkits for recognizing non-classical HLA Class I binders. To fill in this gap, we have developed a deep learning-based tool called DeepHLAPred. The DeepHLAPred used electron-ion interaction pseudo potential, integer numerical mapping and accumulated amino acid frequency as initial representation of non-classical HLA binder sequence. The deep learning module was used to further refine high-level representations. The deep learning module comprised two parallel convolutional neural networks, each followed by maximum pooling layer, dropout layer, and bi-directional long short-term memory network. The experimental results showed that the DeepHLAPred reached the state-of-the-art performanceson the cross-validation test and the independent test. The extensive test demonstrated the rationality of the DeepHLAPred. We further analyzed sequence pattern of non-classical HLA class I binders by information entropy. The information entropy of non-classical HLA binder sequence implied sequence pattern to a certain extent. In addition, we have developed a user-friendly webserver for convenient use, which is available at http://www.biolscience.cn/DeepHLApred/ . The tool and the analysis is helpful to detect non-classical HLA Class I binder. The source code and data is available at https://github.com/tangxingyu0/DeepHLApred .

The significance of low-quality data in unlabeled medical images is always underestimated. We believe that these underestimated data contain valuable information that remains largely unexplored. We present a novel uncertainty-guided different levels of pseudo-labels (UDLP) framework to explore the underestimated data in medical images. The framework consists of a student-teacher model that uses uncertainty to classify the pseudo-labels predicted by the teacher model into three levels: high confidence, low confidence and unreliability. The student model learns directly from high-confidence pseudo-labels. By using the confident learning method in low-confidence pseudo-labels, the teacher model corrects the noisy labels in low-confidence voxels to provide positive feature information for the student model. We design a method for removing unreliable pseudo-labels, to further enhance model’s generalizability. The proposed framework UDLP is evaluated on two datasets and demonstrates superior performance compared to other state-of-the-art methods.