Abstract Accurate prediction of effector proteins secreted by Gram-negative bacteria is important for elucidating bacterial pathogenic mechanisms and developing precise anti-infective strategies. Although existing methods have benefited from the strong sequence feature extraction capacity of pretrained protein language models, reliance on linear sequence information alone often fails to fully capture the three-dimensional conformational signals required for virulence functions. Meanwhile, conventional structure-based methods are limited by the scarcity of experimentally resolved protein structures. To address these challenges, We propose GeoEPred, a multimodal deep learning framework designed for the synergistic modeling of protein sequence and structure to identify Gram-negative bacterial effector proteins. Specifically, the model integrates sequence-contextual embeddings from a pretrained protein language model with three-dimensional structural representations predicted by ESMFold. A feature projection network refines fine-grained sequence signals associated with effector functions, while geometric vector perceptrons characterize inter-residue orientations, distances, and local spatial topology to capture potential structural conformational motifs. To further enable effective cross-modal fusion, we design a cross-modal alignment and feature-tokenized self-attention module. This module enhances consistency between the sequence-semantic and structural-geometric spaces through contrastive learning and models associations between linear functional motifs and spatial conformational patterns at a fine-grained token level. Extensive evaluations on multiple benchmark datasets show that GeoEPred achieves better predictive performance than existing leading models in T3SE, T4SE, and T6SE prediction tasks, while maintaining stable performance in remote homolog recognition scenarios. Moreover, the modular and extensible architecture of GeoEPred demonstrates strong generalization ability and substantial application potential for genome-scale effector protein discovery. Author summary Secreted effector proteins are central virulence factors used by many Gram-negative bacterial pathogens to execute infection strategies. Their functions are governed not only by secretion signals and short linear motifs in the amino acid sequence, but also by three-dimensional folds, local domains, and surface geometric patterns. However, current predictors mainly exploit sequence-contextual features, limiting their ability to model the correspondence between linear sequence signals and spatial conformational motifs, and thereby constraining accuracy and interpretability. Here, we present GeoEPred, a multimodal deep learning framework for secreted effector protein identification. GeoEPred couples sequence-semantic embeddings from a pretrained protein language model with structural representations learned by geometric vector perceptrons. A cross-modal alignment and interaction module uses contrastive learning to improve functional consistency between sequence and structure modalities, while feature-token attention captures fine-grained links between key linear and conformational motifs. Across benchmark datasets covering multiple effector types, GeoEPred outperforms existing state-of-the-art methods and provides interpretable evidence from sequence fragments, structural regions, and cross-modal associations, supporting functional annotation, pathogenic mechanism analysis, and experimental validation.

Identifying effector proteins of secretion systems in Gram-negative bacteria is crucial for deciphering their pathogenic mechanisms and guiding the development of antimicrobial strategies. Extracting evolutionary and sequence features using pre-trained protein language models (PLMs) has emerged as an effective approach to improve the performance of effector protein prediction. However, the high-dimensional features generated by PLMs contain extensive general biological information, making it difficult to focus on core features when applied directly to effector protein tasks, which in turn limits prediction performance. In this study, we propose MoCETSE, a deep learning model for predicting effector proteins in Gram-negative bacteria. Specifically, MoCETSE first extracts contextual representations of sequences using the pre-trained protein language model ESM-1b. Subsequently, it refines key functional features via a target preprocessing network to construct more expressive sequence representations. Finally, integrated with a transformer module incorporating relative positional encoding, MoCETSE explicitly models the relative spatial relationships between residues, enabling highly accurate prediction of secreted effector proteins. MoCETSE exhibits excellent and robust performance in both five-fold cross-validation and independent testing. Benchmark results demonstrate that it maintains strong competitiveness compared to existing binary and multi-class predictors. Additionally, the model can effectively perform genome-wide effector protein prediction, showing outstanding specificity and reliability. MoCETSE provides an efficient and robust computational framework for the accurate identification of bacterial effector substrates and offers key biological insights.

Catalysts accelerate chemical processes by lowering activation energy through alternative reaction pathways. However, existing prediction methods often struggle to capture intricate molecular information and overlook critical structural details such as atomic composition and bond topology. To address these challenges, we propose CataCon, a deep neural network framework grounded in contrastive learning. The model initially utilizes GraphSAGE to generate robust molecular graph embeddings for reactants, products, and candidate catalysts, thereby capturing their structural features comprehensively. Subsequently, a contrastive learning module constructs positive and negative sample pairs to align feature representations between reactant-product combinations and catalysts. This alignment enhances the model's capacity to identify potential interactions. Experimental results on public catalytic reaction datasets demonstrate that CataCon significantly outperforms baseline methods in catalyst classification tasks. Ablation studies further confirm the efficacy of combining graph representation learning with contrastive strategies for integrating multidimensional molecular information. Moreover, t-SNE visualization and interpretability analyses elucidate the underlying decision mechanisms of the model and provide novel perspectives for understanding the complex relationships among reaction components. Scientific contribution This study introduces CataCon, a novel contrastive graph representation learning framework designed to predict optimal catalysts for chemical reactions. This approach overcomes the limitations of existing methods that rely on simplistic catalyst labels by generating rich structural embeddings for all reaction components and aligning reaction and catalyst features through contrastive learning. By providing a powerful tool for the rational screening of catalyst candidates, CataCon achieves superior accuracy and has the potential to significantly accelerate materials discovery and optimize chemical synthesis.

Protein-peptide interactions are essential to cellular processes and disease mechanisms. Identifying protein-peptide binding residues is critical for understanding peptide function and advancing drug discovery. However, experimental methods are costly and time-intensive, while existing computational approaches often predict interactions or binding residues separately, lack effective feature integration, or rely heavily on limited high-quality structural data. To address these challenges, we propose UMPPI (Unveiling Multilevel Protein-Peptide Interaction), a multiobjective framework based on the pretrained protein language model ESM2. UMPPI simultaneously predicts binary protein-peptide interactions and binding residues on both peptides and proteins through a multiobjective optimization strategy. By integrating ESM2 to encode sequences and extract latent structural information, UMPPI bridges the gap between sequence-based and structure-based methods. Extensive experiments demonstrated that UMPPI successfully captured binary interactions between peptides and proteins and identified the binding residues on peptides and proteins. UMPPI can serve as a useful tool for protein-peptide interaction prediction and identification of critical binding residues, thereby facilitating the peptide drug discovery process.

DNA methylation is an epigenetic modification that plays a crucial role in genome stability and cellular specialization, essential for maintaining normal cellular function and development, also a manifestation indicator of some diseases. Various tools have been proposed for methylation detection, typically leveraging a third-generation sequencing technology called nanopore sequencing, which provides more accurate DNA sequencing data. However, existing tools have their own limitations and advantages in terms of computational resources and information processing, without achieving a good balance. In this situation, we developed EDNTOM (Ensemble Deep Network Tool Of Methylation), a DNA methylation detection tool based on deep learning technology. We employed ensemble learning techniques, integrating predictions from multiple pretrained single models, and introduced an attention weight mechanism to provide accurate and reliable detection, reducing the consumption of computational resources. Results demonstrate that EDNTOM outperforms individual models. Additionally, in cross-species transfer experiments, EDNTOM exhibits strong transfer learning capabilities. We hope this work can provide a more powerful and reliable solution for methylation detection, contributing to the fields of biological science and medicine. The project code is available at https://github.com/ViceMusic/EDNTOM.

Accurate identification of N7-methylguanosine (m7G) modification sites plays a critical role in uncovering the regulatory mechanisms of various biological processes, including human development, tumor initiation, and progression. However, existing prediction methods still suffer from limited representational power, redundant feature fusion, insufficient utilization of biological prior knowledge, and poor interpretability. In this study, we propose a novel deep learning model named MCAMEF-BERT. This model adopts a parallel architecture that integrates both a DNABERT-2-based pretrained model branch and multiple traditional feature encoding branches, enabling comprehensive multi-perspective sequence feature extraction. To address the redundancy issue in feature fusion, we introduce a multi-channel attention module. Our model demonstrates superior accuracy and effectiveness on datasets from m7GHub, outperforming other state-of-the-art classifiers. Furthermore, we validate the interpretability of MCAMEF-BERT through in silico saturation mutagenesis experiments, and confirm its robustness in motif recognition. Moreover, its generalization capability is validated across diverse RNA modification site prediction tasks.

RNA modification, as a crucial post-transcriptional regulatory mechanism, plays a pivotal role in normal physiological processes and is closely associated with the onset and progression of various human diseases. Recent studies have highlighted significant alterations in the level of RNA modifications, including m6A, m6Am, m1A, m5C, m7G, ac4C, Ψ, and A-to-I editing, across multiple diseases. These findings suggest the potential of RNA modifications and their regulatory factors as biomarkers for early disease diagnosis and prognosis. This review provides an overview of statistical methods, machine learning techniques employed in identifying disease diagnostic and prognostic biomarkers, along with relevant evaluation metrics and bioinformatics tools. We further explore the types of common RNA modifications, the modifying proteins involved, and the underlying mechanisms of modification. The focus of this paper is on the application of machine learning algorithms in discovering RNA modification-related biomarkers, particularly for disease diagnosis and prognosis. By reviewing recent advancements in the identification of disease biomarkers, and analyzing the prospects and challenges of their clinical application, we aim to offer insights into the mining methods of RNA modifications and their associated factors as disease diagnostic or prognostic biomarkers, providing a valuable reference for future research and clinical practice.

Drug-drug interactions (DDIs) are critical in pharmaceutical research, as adverse interactions can pose significant risks for patient treatment plans. Accurate prediction of DDI events risk levels can provide valuable guidance for designing safer and more effective medical regimens. However, existing approaches often focus on interaction networks while overlooking the inherent molecular properties of drugs. In this study, we present Taco-DDI, a novel drug representation learning framework that utilizes a graph transformer-based model combined with a dynamic co-attention mechanism. Taco-DDI leverages the transformer architecture to derive atom-level feature encodings, capturing comprehensive molecular representations. Furthermore, it employs an adaptive co-attention matrix to identify essential substructures in drug molecules solely from structural information. Our results demonstrate that Taco-DDI achieves a 6.59 % relative accuracy improvement in DDI events risk levels prediction. Additionally, interpretability analysis confirms that Taco-DDI provides meaningful insights into DDI mechanisms, highlighting its practical utility as a robust tool for identifying DDI events risk levels.

Recent advancements in spatial transcriptomics (ST) technology have generated substantial volumes of spatial transcriptome data. However, the quality of this data is often compromised due to the limitations of current sequencing technologies. To address this issue, DiffusionST proposes a method for imputing ST data and clustering the imputed data. The method employs a graph convolutional network model combined with a newly designed loss function, denoising data using the zero-inflated negative binomial distribution, and data enhancement through a diffusion model to improve clustering accuracy. DiffusionST demonstrates superior clustering accuracy compared to eight of the most popular ST clustering algorithms. DiffusionST also excels in data imputation when compared to five single-cell RNA sequencing imputation algorithms. Additionally, DiffusionST's robustness against noise is quantitatively validated by manually introducing random dropout noise into the dataset, where our model significantly enhances the quality of ST data. Moreover, DiffusionST is well-suited for high-resolution ST data and has been demonstrated, through survival analysis and cell-cell communication studies, to dissect spatial domains within breast cancer tissues. These findings provide strong evidence of DiffusionST's efficacy in handling ST data especially with strong noise, making it a valuable tool in this field.

Spatial transcriptomics maps gene expression across tissues, yet data sparsity and noise challenge long-range dependency modeling, limiting accurate spatial domain delineation. In this study, we present TOGAR, a token-gated generative refinement model that unifies denoising, spatial enhancement, and clustering for spatial transcriptomics. Firstly, the model combines a graph convolutional network loss with a loss based on the zero-inflated negative binomial distribution to reduce noise and enhance signal clarity in sparse count data. It then employs a UGate-based diffusion backbone, which integrates token gating, gated linear attention, and rotary positional embedding for generative spatial refinement. Finally, similarity-guided averaging along diffusion trajectories provides stable spot-level estimates, and clustering of the refined representations produces spatial domains with sharp boundaries suitable for downstream analyses. We evaluate TOGAR across three spatial transcriptomics platforms. In benchmarks on twelve slices against seven popular methods, TOGAR consistently achieves or exceeds clustering accuracy, demonstrating superior stability. TOGAR effectively recovers coherent cortical layer organization, delineates fine-grained tumor subdomains associated with immune activity and extracellular matrix remodeling, and generates clearer, biologically interpretable domain boundaries. Notably, TOGAR excels in detecting extremely small and rare spatial structures, successfully identifying biologically important regions that other methods completely miss, while maintaining boundary integrity in complex multi-cluster structures and avoiding issues of over-connectivity or incomplete detection.

Peptide precursors, as the source molecules of bioactive peptides, play essential roles in neuroregulation, immune defense, and drug development. Their accurate identification is crucial for elucidating mechanisms of life regulation and developing novel therapeutics. However, the complexity and diversity of peptide precursor sequences pose significant challenges to prediction tasks. Existing methods predominantly rely on sequence features or structural features, hindering the full exploitation of complementary information between modalities and consequently limiting prediction performance. We introduce ProjFusNet, a deep learning framework that integrates evolutionary-scale protein sequence representations from ESM-2 with structural features via a projected multimodal fusion strategy. A bidirectional LSTM is further employed to model the complex interactions between sequence and structure. In rigorous five-fold cross-validation, ProjFusNet demonstrates improved performance across key metrics, including ACC, SN, AUC, SP, and MCC, compared to single-feature models.

Neuropeptides are key signaling molecules that regulate fundamental physiological processes ranging from metabolism to cognitive function. However, accurate identification is a huge challenge due to sequence heterogeneity, obscured functional motifs and limited experimentally validated data. Accurate identification of neuropeptides is critical for advancing neurological disease therapeutics and peptide-based drug design. Existing neuropeptide identification methods rely on manual features combined with traditional machine learning methods, which are difficult to capture the deep patterns of sequences. To address these limitations, we propose NeuroPred-AIMP (adaptive integrated multimodal predictor), an interpretable model that synergizes global semantic representation of the protein language model (ESM) and the multiscale structural features of the temporal convolutional network (TCN). The model introduced the adaptive features fusion mechanism of residual enhancement to dynamically recalibrate feature contributions, to achieve robust integration of evolutionary and local sequence information. The experimental results demonstrated that the proposed model showed excellent comprehensive performance on the independence test set, with an accuracy of 92.3% and the AUROC of 0.974. Simultaneously, the model showed good balance in the ability to identify positive and negative samples, with a sensitivity of 92.6% and a specificity of 92.1%, with a difference of less than 0.5%. The result fully confirms the effectiveness of the multimodal features strategy in the task of neuropeptide recognition.

RNA modifications play a crucial role in enhancing the structural and functional diversity of RNA molecules and regulating various stages of the RNA life cycle. Among these modifications, N6-Methyladenosine (m6A) is the most common internal modification in eukaryotic mRNAs and has been extensively studied over the past decade. Accurate identification of m6A modification sites is essential for understanding their function and underlying mechanisms. Traditional methods predominantly rely on machine learning techniques to recognize m6A sites, which often fail to capture the contextual features of these sites comprehensively. In this study, we comprehensively summarize previously published methods based on machine learning and deep learning. We also validate multiple deep learning approaches on benchmark dataset, including previously underutilized methods in m6A site prediction, pre-trained models specifically designed for biological sequence and other basic deep learning methods. Additionally, we further analyze the dataset features and interpret the model's predictions to enhance understanding. Our experimental results clearly demonstrate the effectiveness of the deep learning models, elucidating their strong potential in accurately recognizing m6A modification sites.

Background: Axillary lymph node (ALN) status is a crucial prognostic indicator for breast cancer metastasis, with manual interpretation of whole slide images (WSIs) being the current standard practice. However, this method is subjective and time-consuming. Recent advancements in deep learning-based methods for medical image analysis have shown promise in improving clinical diagnosis. This study aims to leverage these technological advancements to develop a deep learning model based on features extracted from primary tumor biopsies for preoperatively identifying ALN metastasis in early-stage breast cancer patients with negative nodes. Methods: We present DLCNBC-SA, a deep learning-based network specifically tailored for core needle biopsy and clinical data feature extraction, which integrates a self-attention mechanism (CNBC-SA). The proposed model consists of a feature extractor based on convolutional neural network (CNN) and an improved self-attention mechanism module, which can preserve the independence of features in WSIs for analysis and enhancement to provide rich feature representation. To validate the performance of the proposed model, we conducted comparative experiments and ablation studies using publicly available datasets, and verification was performed through quantitative analysis. Results: The comparative experiment illustrates the superior performance of the proposed model in the task of binary classification of ALNs, as compared to alternative methods. Our method achieved outstanding performance [area under the curve (AUC): 0.882] in this task, significantly surpassing the state-of-the-art (SOTA) method on the same dataset (AUC: 0.862). The ablation experiment reveals that incorporating RandomRotation data augmentation technology and utilizing Adadelta optimizer can effectively enhance the performance of the proposed model. Conclusions: The experimental results demonstrate that the model proposed in this paper outperforms the SOTA model on the same dataset, thereby establishing its reliability as an assistant for pathologists in analyzing WSIs of breast cancer. Consequently, it significantly enhances both the efficiency and accuracy of doctors during the diagnostic process.

Face recognition has a wide range of applications in daily life, and it is of great practical significance to improve the accuracy of face recognition. The current 2D face recognition technology has an accuracy rate close to 99% under ideal conditions, but its performance is still affected by factors like lighting changes and occlusion. The rapid development of 3D face reconstruction technology has greatly enhanced the ability to create 3D faces from 2D images. By parameterizing the face model for 3D reconstruction, an occlusion-free 3D face can be generated, thus effectively addressing the challenges in 2D recognition. To address the above problems, this paper proposes a 3D face recognition method based on the DECA model. By combining the 3D face shape maps produced by the DECA model with the traditional 2D color maps, the accuracy of feature extraction is improved. We evaluated our method through closed-set testing using the Texas3DFRD dataset. The experimental results show that using both 2D color maps and 3D reconstruction-generated shape maps as input for feature extraction can significantly improve the accuracy of face matching and recognition, verifying the advantages of 3D face reconstruction technique in recognition tasks.

MOTIVATION:5-Methylcytosine (5mC), a fundamental element of DNA methylation in eukaryotes, plays a vital role in gene expression regulation, embryonic development, and other biological processes. Although several computational methods have been proposed for detecting the base modifications in DNA like 5mC sites from Nanopore sequencing data, they face challenges including sensitivity to noise, and ignoring the imbalanced distribution of methylation sites in real-world scenarios.RESULTS:Here, we develop NanoCon, a deep hybrid network coupled with contrastive learning strategy to detect 5mC methylation sites from Nanopore reads. In particular, we adopted a contrastive learning module to alleviate the issues caused by imbalanced data distribution in nanopore sequencing, offering a more accurate and robust detection of 5mC sites. Evaluation results demonstrate that NanoCon outperforms existing methods, highlighting its potential as a valuable tool in genomic sequencing and methylation prediction. In addition, we also verified the effectiveness of our representation learning ability on two datasets by visualizing the dimension reduction of the features of methylation and nonmethylation sites from our NanoCon. Furthermore, cross-species and cross-5mC methylation motifs experiments indicated the robustness and the ability to perform transfer learning of our model. We hope this work can contribute to the community by providing a powerful and reliable solution for 5mC site detection in genomic studies.AVAILABILITY AND IMPLEMENTATION:The project code is available at https://github.com/Challis-yin/NanoCon.

In recent years, optical flow and transformer have been applied to video inpainting successfully. However, in the previous works, the networks for extracting and completing optical flow need to be trained separately, which is a cumbersome process. Meanwhile, the transformer architecture suffers from high computational complexity and imprecise attention retrieval. Therefore, a video inpaiting model by fusing optical flows with a spatial-temporal transformer is proposed in this paper. In this model, a novel cross-stacked spatial-temporal transformer is designed, utilizing distant frames as reference frames within the temporal transformer. Simultaneously, motion information from optical flows is integrated into the spatial transformer to refine the attention retrieval. Within these transformer blocks, a window focal attention mechanism is introduced to achieve a combination of coarse-grained and fine-grained attention. This model maintains inpainting quality while lowering computational complexity effectively. Additionally, by training the flow completion network and other modules jointly, an end-to-end inpainting process is achieved. The effectiveness and superiority of the proposed model are confirmed by both quantitative and qualitative evaluation results.

It is very difficult to accomplish the 3D reconstruction of the clothed human body from a single RGB image, because the 2D image lacks the represen-tation information of the 3D human body, especially for the clothed human body. In order to solve this problem, we introduced a priority scheme of different body parts spatial information and proposed PointHuman network. PointHuman com-bines the spatial feature of the parametric model of the human body with the implicit functions without expressive restrictions. In PointHuman reconstruction framework, we use Point Transformer to extract the semantic spatial feature of the parametric model of the human body to regularize the implicit function of the neural network, which extends the generalization ability of the neural network to complex human poses and various styles of clothing. Moreover, considering the ambiguity of depth information, we estimate the depth of the parameterized model after point cloudization, and obtain an offset depth value. The offset depth value improves the consistency between the parameterized model and the neural implicit function, and accuracy of human reconstruction models. Finally, we opti-mize the restoration of the parametric model from a single image, and propose a depth perception method. This method further improves the estimation ac-curacy of the parametric model and finally improves the effectiveness of human reconstruction. Our method achieves competitive performance on the THuman dataset.

Prediction of essential genes in a life organism is one of the central tasks in synthetic biology. Computational predictors are desired because experimental data is often unavailable. Recently, some sequence-based predictors have been constructed to identify essential genes. However, their predictive performance should be further improved. One key problem is how to effectively extract the sequence-based features, which are able to discriminate the essential genes. Another problem is the imbalanced training set. The amount of essential genes in human cell lines is lower than that of non-essential genes. Therefore, predictors trained with such imbalanced training set tend to identify an unseen sequence as a non-essential gene. Here, a new over-sampling strategy was proposed called Clustering based Synthetic Minority Oversampling Technique (CSMOTE) to overcome the imbalanced data issue. Combining CSMOTE with the Z curve, the global features, and Support Vector Machines, a new protocol called iEsGene-CSMOTE was proposed to identify essential genes. The rigorous jackknife cross validation results indicated that iEsGene-CSMOTE is better than the other competing methods. The proposed method outperformed λ-interval Z curve by 35.48% and 11.25% in terms of Sn and BACC, respectively.

Enhancers are crucial for precise regulation of gene expression, while enhancer identification and strength prediction are challenging because of their free distribution and tremendous number of similar fractions in the genome. Although several bioinformatics tools have been developed, shortfalls in these models remain, and their performances need further improvement. In the present study, a two-layer predictor called Enhancer-FRL was proposed for identifying enhancers (enhancers or nonenhancers) and their activities (strong and weak). More specifically, to build an efficient model, the feature representation learning scheme was applied to generate a 50D probabilistic vector based on 10 feature encodings and five machine learning algorithms. Subsequently, the multiview probabilistic features were integrated to construct the final prediction model. Compared with the single feature-based model, Enhancer-FRL showed significant performance improvement and model robustness. Performance assessment on the independent test dataset indicated that the proposed model outperformed state-of-the-art available toolkits. The webserver Enhancer-FRL is freely accessible at http://lab.malab.cn/∼wangchao/softwares/Enhancer-FRL/, The code and datasets can be downloaded at the webserver page or at the Github https://github.com/wangchao-malab/Enhancer-FRL/.