ERT-GFAN: A multimodal drug-target interaction prediction model based on molecular biology and knowledge-enhanced attention mechanism.

In drug discovery, precisely identifying drug-target interactions is crucial for finding new drugs and understanding drug mechanisms. Evolving drug/target heterogeneous data presents challenges in obtaining multimodal representation in drug-target prediction(DTI). To deal with this, we propose 'ERT-GFAN', a multimodal drug-target interaction prediction model inspired by molecular biology. Firstly, it integrates bio-inspired principles to obtain structure feature of drugs and targets using Extended Connectivity Fingerprints(ECFP). Simultaneously, the knowledge graph embedding model RotatE is employed to discover the interaction feature of drug-target pairs. Subsequently, Transformer is utilized to refine the contextual neighborhood features from the obtained structure feature and interaction features, and multi-modal high-dimensional fusion features of the three-modal information constructed. Finally, the final DTI prediction results are outputted by integrating the multimodal fusion features into a graphical high-dimensional fusion feature attention network (GFAN) using our innovative multimodal high-dimensional fusion feature attention. This multimodal approach offers a comprehensive understanding of drug-target interactions, addressing challenges in complex knowledge graphs. By combining structure feature, interaction feature, and contextual neighborhood features, 'ERT-GFAN' excels in predicting DTI. Empirical evaluations on three datasets demonstrate our method's superior performance, with AUC of 0.9739, 0.9862, and 0.9667, AUPR of 0.9598, 0.9789, and 0.9750, and Mean Reciprocal Rank(MRR) of 0.7386, 0.7035, and 0.7133. Ablation studies show over a 5% improvement in predictive performance compared to baseline unimodal and bimodal models. These results, along with detailed case studies, highlight the efficacy and robustness of our approach.

Drug repositioning based on residual attention network and free multiscale adversarial training.

BACKGROUND: Conducting traditional wet experiments to guide drug development is an expensive, time-consuming and risky process. Analyzing drug function and repositioning plays a key role in identifying new therapeutic potential of approved drugs and discovering therapeutic approaches for untreated diseases. Exploring drug-disease associations has far-reaching implications for identifying disease pathogenesis and treatment. However, reliable detection of drug-disease relationships via traditional methods is costly and slow. Therefore, investigations into computational methods for predicting drug-disease associations are currently needed. RESULTS: This paper presents a novel drug-disease association prediction method, RAFGAE. First, RAFGAE integrates known associations between diseases and drugs into a bipartite network. Second, RAFGAE designs the Re_GAT framework, which includes multilayer graph attention networks (GATs) and two residual networks. The multilayer GATs are utilized for learning the node embeddings, which is achieved by aggregating information from multihop neighbors. The two residual networks are used to alleviate the deep network oversmoothing problem, and an attention mechanism is introduced to combine the node embeddings from different attention layers. Third, two graph autoencoders (GAEs) with collaborative training are constructed to simulate label propagation to predict potential associations. On this basis, free multiscale adversarial training (FMAT) is introduced. FMAT enhances node feature quality through small gradient adversarial perturbation iterations, improving the prediction performance. Finally, tenfold cross-validations on two benchmark datasets show that RAFGAE outperforms current methods. In addition, case studies have confirmed that RAFGAE can detect novel drug-disease associations. CONCLUSIONS: The comprehensive experimental results validate the utility and accuracy of RAFGAE. We believe that this method may serve as an excellent predictor for identifying unobserved disease-drug associations.

An unsupervised multi-view contrastive learning framework with attention-based reranking strategy for entity alignment.

Entity alignment is a crucial task in knowledge graphs, aiming to match corresponding entities from different knowledge graphs. Due to the scarcity of pre-aligned entities in real-world scenarios, research focused on unsupervised entity alignment has become more popular. However, current unsupervised entity alignment methods suffer from a lack of informative entity guidance, hindering their ability to accurately predict challenging entities with similar names and structures. To solve these problems, we present an unsupervised multi-view contrastive learning framework with an attention-based reranking strategy for entity alignment, named AR-Align. In AR-Align, two kinds of data augmentation methods are employed to provide a complementary view for neighborhood and attribute, respectively. Next, a multi-view contrastive learning method is introduced to reduce the semantic gap between different views of the augmented entities. Moreover, an attention-based reranking strategy is proposed to rerank the hard entities through calculating their weighted sum of embedding similarities on different structures. Experimental results indicate that AR-Align outperforms most both supervised and unsupervised state-of-the-art methods on three benchmark datasets.

Air quality forecasting using a spatiotemporal hybrid deep learning model based on VMD-GAT-BiLSTM.

The precise forecasting of air quality is of great significance as an integral component of early warning systems. This remains a formidable challenge owing to the limited information of emission source and the considerable uncertainties inherent in dynamic processes. To improve the accuracy of air quality forecasting, this work proposes a new spatiotemporal hybrid deep learning model based on variational mode decomposition (VMD), graph attention networks (GAT) and bi-directional long short-term memory (BiLSTM), referred to as VMD-GAT-BiLSTM, for air quality forecasting. The proposed model initially employ a VMD to decompose original PM2.5 data into a series of relatively stable sub-sequences, thus reducing the influence of unknown factors on model prediction capabilities. For each sub-sequence, a GAT is then designed to explore deep spatial relationships among different monitoring stations. Next, a BiLSTM is utilized to learn the temporal features of each decomposed sub-sequence. Finally, the forecasting results of each decomposed sub-sequence are aggregated and summed as the final air quality prediction results. Experiment results on the collected Beijing air quality dataset show that the proposed model presents superior performance to other used methods on both short-term and long-term air quality forecasting tasks.

DKPE-GraphSYN: a drug synergy prediction model based on joint dual kernel density estimation and positional encoding for graph representation.

Introduction: Synergistic medication, a crucial therapeutic strategy in cancer treatment, involves combining multiple drugs to enhance therapeutic effectiveness and mitigate side effects. Current research predominantly employs deep learning models for extracting features from cell line and cancer drug structure data. However, these methods often overlook the intricate nonlinear relationships within the data, neglecting the distribution characteristics and weighted probability densities of gene expression data in multi-dimensional space. It also fails to fully exploit the structural information of cancer drugs and the potential interactions between drug molecules. Methods: To overcome these challenges, we introduce an innovative end-to-end learning model specifically tailored for cancer drugs, named Dual Kernel Density and Positional Encoding (DKPE) for Graph Synergy Representation Network (DKPEGraphSYN). This model is engineered to refine the prediction of drug combination synergy effects in cancer. DKPE-GraphSYN utilizes Dual Kernel Density Estimation and Positional Encoding techniques to effectively capture the weighted probability density and spatial distribution information of gene expression, while exploring the interactions and potential relationships between cancer drug molecules via a graph neural network. Results: Experimental results show that our prediction model achieves significant performance enhancements in forecasting drug synergy effects on a comprehensive cancer drug and cell line synergy dataset, achieving an AUPR of 0.969 and an AUC of 0.976. Discussion: These results confirm our model's superior accuracy in predicting cancer drug combinations, providing a supportive method for clinical medication strategy in cancer.

Integrating spatial transcriptomics and bulk RNA-seq: predicting gene expression with enhanced resolution through graph attention networks.

Spatial transcriptomics data play a crucial role in cancer research, providing a nuanced understanding of the spatial organization of gene expression within tumor tissues. Unraveling the spatial dynamics of gene expression can unveil key insights into tumor heterogeneity and aid in identifying potential therapeutic targets. However, in many large-scale cancer studies, spatial transcriptomics data are limited, with bulk RNA-seq and corresponding Whole Slide Image (WSI) data being more common (e.g. TCGA project). To address this gap, there is a critical need to develop methodologies that can estimate gene expression at near-cell (spot) level resolution from existing WSI and bulk RNA-seq data. This approach is essential for reanalyzing expansive cohort studies and uncovering novel biomarkers that have been overlooked in the initial assessments. In this study, we present STGAT (Spatial Transcriptomics Graph Attention Network), a novel approach leveraging Graph Attention Networks (GAT) to discern spatial dependencies among spots. Trained on spatial transcriptomics data, STGAT is designed to estimate gene expression profiles at spot-level resolution and predict whether each spot represents tumor or non-tumor tissue, especially in patient samples where only WSI and bulk RNA-seq data are available. Comprehensive tests on two breast cancer spatial transcriptomics datasets demonstrated that STGAT outperformed existing methods in accurately predicting gene expression. Further analyses using the TCGA breast cancer dataset revealed that gene expression estimated from tumor-only spots (predicted by STGAT) provides more accurate molecular signatures for breast cancer sub-type and tumor stage prediction, and also leading to improved patient survival and disease-free analysis. Availability: Code is available at https://github.com/compbiolabucf/STGAT.

Enhanced botnet detection in IoT networks using zebra optimization and dual-channel GAN classification.

The Internet of Things (IoT) permeates various sectors, including healthcare, smart cities, and agriculture, alongside critical infrastructure management. However, its susceptibility to malware due to limited processing power and security protocols poses significant challenges. Traditional antimalware solutions fall short in combating evolving threats. To address this, the research work developed a feature selection-based classification model. At first stage, a preprocessing stage enhances dataset quality through data smoothing and consistency improvement. Feature selection via the Zebra Optimization Algorithm (ZOA) reduces dimensionality, while a classification phase integrates the Graph Attention Network (GAN), specifically the Dual-channel GAN (DGAN). DGAN incorporates Node Attention Networks and Semantic Attention Networks to capture intricate IoT device interactions and detect anomalous behaviors like botnet activity. The model's accuracy is further boosted by leveraging both structural and semantic data with the Sooty Tern Optimization Algorithm (STOA) for hyperparameter tuning. The proposed STOA-DGAN model achieves an impressive 99.87% accuracy in botnet activity classification, showcasing robustness and reliability compared to existing approaches.

Identifying circRNA-disease association based on relational graph attention network and hypergraph attention network.

In recent years, with the in-depth study of circRNA, scholars have begun to discover a synergistic relationship between circRNA and microorganisms. Traditional wet lab experiments in biology require expensive financial, material, and human resources to investigate the relationship between circRNA and diseases. Therefore, we propose a new predictive model for inferring the association between circRNA and diseases, called HAGACDA. Specifically, we first aggregate the unique features of circRNA and diseases themselves through singular value decomposition, Pearson similarity, and the biological information characteristics of circRNA and diseases. Utilizing the competitive relationships between miRNA and other microorganisms, we construct a circRNA-miRNA-disease multi-source heterogeneous network. Subsequently, we use a relational graph attention network to aggregate features based on the structural connections between different nodes. To address the inherent limitations in capturing high-order patterns in edge sets, we integrate a hypergraph attention network to extract features of circRNA and diseases. Finally, association prediction scores for node pairs are obtained through a multilayer perceptron. We conducted a comprehensive analysis of the model, including comparative experiments and case studies. Experimental results demonstrate that our model accurately predicts the association between circRNA and diseases.

Predicting bond dissociation energies of cyclic hypervalent halogen reagents using DFT calculations and graph attention network model.

Although hypervalent iodine(III) reagents have become staples in organic chemistry, the exploration of their isoelectronic counterparts, namely hypervalent bromine(III) and chlorine(III) reagents, has been relatively limited, partly due to challenges in synthesizing and stabilizing these compounds. In this study, we conduct a thorough examination of both homolytic and heterolytic bond dissociation energies (BDEs) critical for assessing the chemical stability and functional group transfer capability of cyclic hypervalent halogen compounds using density functional theory (DFT) analysis. A moderate linear correlation was observed between the homolytic BDEs across different halogen centers, while a strong linear correlation was noted among the heterolytic BDEs across these centers. Furthermore, we developed a predictive model for both homolytic and heterolytic BDEs of cyclic hypervalent halogen compounds using machine learning algorithms. The results of this study could aid in estimating the chemical stability and functional group transfer capabilities of hypervalent bromine(III) and chlorine(III) reagents, thereby facilitating their development.

A multi-feature spatial-temporal fusion network for traffic flow prediction.

The traffic flow prediction is the key to alleviate traffic congestion, yet very challenging due to the complex influence factors. Currently, the most of deep learning models are designed to dig out the intricate dependency in continuous standardized sequences, which are dependent to high requirements for data continuity and regularized distribution. However, the data discontinuity and irregular distribution are inevitable in the real-world practical application, then we need find a way to utilize the powerful effect of the multi-feature fusion rather than continuous relation in standardized sequences. To this end, we conduct the prediction based on the multiple traffic features reflecting the complex influence factors. Firstly, we propose the ATFEM, an adaptive traffic features extraction mechanism, which can select important influence factors to construct joint temporal features matrix and global spatial features matrix according to the traffic condition. In this way, the feature's representation ability can be improved. Secondly, we propose the MFSTN, a multi-feature spatial-temporal fusion network, which include the temporal transformer encoder and graph attention network to obtain the latent representation of spatial-temporal features. Especially, we design the scaled spatial-temporal fusion module, which can automatically learn optimal fusion weights, further adapt to inconsistent spatial-temporal dimensions. Finally, the multi-layer perceptron gets the mapping function between these comprehensive features and traffic flow. This method helps to improve the interpretability of the prediction. Experimental results show that the proposed model outperforms a variety of baselines, and it can accurately predict the traffic flow when the data missing rate is high.

Tissue module discovery in single-cell-resolution spatial transcriptomics data via cell-cell interaction-aware cell embedding.

Computational methods are desired for single-cell-resolution spatial transcriptomics (ST) data analysis to uncover spatial organization principles for how individual cells exert tissue-specific functions. Here, we present ST data analysis via interaction-aware cell embedding (SPACE), a deep-learning method for cell-type identification and tissue module discovery from single-cell-resolution ST data by learning a cell representation that captures its gene expression profile and interactions with its spatial neighbors. SPACE identified spatially informed cell subtypes defined by their special spatial distribution patterns and distinct proximal-interacting cell types. SPACE also automatically discovered "cell communities"-tissue modules with discernible boundaries and a uniform spatial distribution of constituent cell types. For each cell community, SPACE outputs a characteristic proximal cell-cell interaction network associated with physiological processes, which can be used to refine ligand-receptor-based intercellular signaling analyses. We envision that SPACE can be used in large-scale ST projects to understand how proximal cell-cell interactions contribute to emergent biological functions within cell communities. A record of this paper's transparent peer review process is included in the supplemental information.

Accurate Identification of Spatial Domain by Incorporating Global Spatial Proximity and Local Expression Proximity.

Accurate identification of spatial domains is essential in the analysis of spatial transcriptomics data in order to elucidate tissue microenvironments and biological functions. However, existing methods only perform domain segmentation based on local or global spatial relationships between spots, resulting in an underutilization of spatial information. To this end, we propose SECE, a deep learning-based method that captures both local and global relationships among spots and aggregates their information using expression similarity and spatial similarity. We benchmarked SECE against eight state-of-the-art methods on six real spatial transcriptomics datasets spanning four different platforms. SECE consistently outperformed other methods in spatial domain identification accuracy. Moreover, SECE produced spatial embeddings that exhibited clearer patterns in low-dimensional visualizations and facilitated a more accurate trajectory inference.

Prediction of microbe-drug associations based on a modified graph attention variational autoencoder and random forest.

Introduction: The identification of microbe-drug associations can greatly facilitate drug research and development. Traditional methods for screening microbe-drug associations are time-consuming, manpower-intensive, and costly to conduct, so computational methods are a good alternative. However, most of them ignore the combination of abundant sequence, structural information, and microbe-drug network topology. Methods: In this study, we developed a computational framework based on a modified graph attention variational autoencoder (MGAVAEMDA) to infer potential microbedrug associations by combining biological information with the variational autoencoder. In MGAVAEMDA, we first used multiple databases, which include microbial sequences, drug structures, and microbe-drug association databases, to establish two comprehensive feature matrices of microbes and drugs after multiple similarity computations, fusion, smoothing, and thresholding. Then, we employed a combination of variational autoencoder and graph attention to extract low-dimensional feature representations of microbes and drugs. Finally, the lowdimensional feature representation and graphical adjacency matrix were input into the random forest classifier to obtain the microbe-drug association score to identify the potential microbe-drug association. Moreover, in order to correct the model complexity and redundant calculation to improve efficiency, we introduced a modified graph convolutional neural network embedded into the variational autoencoder for computing low dimensional features. Results: The experiment results demonstrate that the prediction performance of MGAVAEMDA is better than the five state-of-the-art methods. For the major measurements (AUC =0.9357, AUPR =0.9378), the relative improvements of MGAVAEMDA compared to the suboptimal methods are 1.76 and 1.47%, respectively. Discussion: We conducted case studies on two drugs and found that more than 85% of the predicted associations have been reported in PubMed. The comprehensive experimental results validated the reliability of our models in accurately inferring potential microbe-drug associations.

MIFAM-DTI: a drug-target interactions predicting model based on multi-source information fusion and attention mechanism.

Accurate identification of potential drug-target pairs is a crucial step in drug development and drug repositioning, which is characterized by the ability of the drug to bind to and modulate the activity of the target molecule, resulting in the desired therapeutic effect. As machine learning and deep learning technologies advance, an increasing number of models are being engaged for the prediction of drug-target interactions. However, there is still a great challenge to improve the accuracy and efficiency of predicting. In this study, we proposed a deep learning method called Multi-source Information Fusion and Attention Mechanism for Drug-Target Interaction (MIFAM-DTI) to predict drug-target interactions. Firstly, the physicochemical property feature vector and the Molecular ACCess System molecular fingerprint feature vector of a drug were extracted based on its SMILES sequence. The dipeptide composition feature vector and the Evolutionary Scale Modeling -1b feature vector of a target were constructed based on its amino acid sequence information. Secondly, the PCA method was employed to reduce the dimensionality of the four feature vectors, and the adjacency matrices were constructed by calculating the cosine similarity. Thirdly, the two feature vectors of each drug were concatenated and the two adjacency matrices were subjected to a logical OR operation. And then they were fed into a model composed of graph attention network and multi-head self-attention to obtain the final drug feature vectors. With the same method, the final target feature vectors were obtained. Finally, these final feature vectors were concatenated, which served as the input to a fully connected layer, resulting in the prediction output. MIFAM-DTI not only integrated multi-source information to capture the drug and target features more comprehensively, but also utilized the graph attention network and multi-head self-attention to autonomously learn attention weights and more comprehensively capture information in sequence data. Experimental results demonstrated that MIFAM-DTI outperformed state-of-the-art methods in terms of AUC and AUPR. Case study results of coenzymes involved in cellular energy metabolism also demonstrated the effectiveness and practicality of MIFAM-DTI. The source code and experimental data for MIFAM-DTI are available at https://github.com/Search-AB/MIFAM-DTI.

A Study of Adjacent Intersection Correlation Based on Temporal Graph Attention Network.

Traffic state classification and relevance calculation at intersections are both difficult problems in traffic control. In this paper, we propose an intersection relevance model based on a temporal graph attention network, which can solve the above two problems at the same time. First, the intersection features and interaction time of the intersections are regarded as input quantities together with the initial labels of the traffic data. Then, they are inputted into the temporal graph attention (TGAT) model to obtain the classification accuracy of the target intersections in four states-free, stable, slow moving, and congested-and the obtained neighbouring intersection weights are used as the correlation between the intersections. Finally, it is validated by VISSIM simulation experiments. In terms of classification accuracy, the TGAT model has a higher classification accuracy than the three traditional classification models and can cope well with the uneven distribution of the number of samples. The information gain algorithm from the information entropy theory was used to derive the average delay as the most influential factor on intersection status. The correlation from the TGAT model positively correlates with traffic flow, making it interpretable. Using this correlation to control the division of subareas improves the road network's operational efficiency more than the traditional correlation model does. This demonstrates the effectiveness of the TGAT model's correlation.

Prediction of systemic lupus erythematosus-related genes based on graph attention network and deep neural network.

Systemic lupus erythematosus (SLE) is an autoimmune disorder intricately linked to genetic factors, with numerous approaches having identified genes linked to its development, diagnosis and prognosis. Despite genome-wide association analysis and gene knockout experiments confirming some genes associated with SLE, there are still numerous potential genes yet to be discovered. The search for relevant genes through biological experiments entails significant financial and human resources. With the advancement of computational technologies like deep learning, we aim to identify SLE-related genes through deep learning methods, thereby narrowing down the scope for biological experimentation. This study introduces SLEDL, a deep learning-based approach that leverages DNN and graph neural networks to effectively identify SLE-related genes by capturing relevant features in the gene interaction network. The above steps transform the identification of SLE related genes into a binary classification problem, ultimately solved through a fully connected layer. The results demonstrate the superiority of SLEDL, achieving higher AUC (0.7274) and AUPR (0.7599), further validated through case studies.

Malignancy pattern analysis of breast ultrasound images using clinical features and a graph convolutional network.

Objective: Early diagnosis of breast cancer can lead to effective treatment, possibly increase long-term survival rates, and improve quality of life. The objective of this study is to present an automated analysis and classification system for breast cancer using clinical markers such as tumor shape, orientation, margin, and surrounding tissue. The novelty and uniqueness of the study lie in the approach of considering medical features based on the diagnosis of radiologists. Methods: Using clinical markers, a graph is generated where each feature is represented by a node, and the connection between them is represented by an edge which is derived through Pearson's correlation method. A graph convolutional network (GCN) model is proposed to classify breast tumors into benign and malignant, using the graph data. Several statistical tests are performed to assess the importance of the proposed features. The performance of the proposed GCN model is improved by experimenting with different layer configurations and hyper-parameter settings. Results: Results show that the proposed model has a 98.73% test accuracy. The performance of the model is compared with a graph attention network, a one-dimensional convolutional neural network, and five transfer learning models, ten machine learning models, and three ensemble learning models. The performance of the model was further assessed with three supplementary breast cancer ultrasound image datasets, where the accuracies are 91.03%, 94.37%, and 89.62% for Dataset A, Dataset B, and Dataset C (combining Dataset A and Dataset B) respectively. Overfitting issues are assessed through k-fold cross-validation. Conclusion: Several variants are utilized to present a more rigorous and fair evaluation of our work, especially the importance of extracting clinically relevant features. Moreover, a GCN model using graph data can be a promising solution for an automated feature-based breast image classification system.

Epileptic Seizure Prediction Using Spatiotemporal Feature Fusion on EEG.

Electroencephalography (EEG) plays a crucial role in epilepsy analysis, and epileptic seizure prediction has significant value for clinical treatment of epilepsy. Currently, prediction methods using Convolutional Neural Network (CNN) primarily focus on local features of EEG, making it challenging to simultaneously capture the spatial and temporal features from multi-channel EEGs to identify the preictal state effectively. In order to extract inherent spatial relationships among multi-channel EEGs while obtaining their temporal correlations, this study proposed an end-to-end model for the prediction of epileptic seizures by incorporating Graph Attention Network (GAT) and Temporal Convolutional Network (TCN). Low-pass filtered EEG signals were fed into the GAT module for EEG spatial feature extraction, and followed by TCN to capture temporal features, allowing the end-to-end model to acquire the spatiotemporal correlations of multi-channel EEGs. The system was evaluated on the publicly available CHB-MIT database, yielding segment-based accuracy of 98.71%, specificity of 98.35%, sensitivity of 99.07%, and F1-score of 98.71%, respectively. Event-based sensitivity of 97.03% and False Positive Rate (FPR) of 0.03/h was also achieved. Experimental results demonstrated this system can achieve superior performance for seizure prediction by leveraging the fusion of EEG spatiotemporal features without the need of feature engineering.

MFSynDCP: multi-source feature collaborative interactive learning for drug combination synergy prediction.

Drug combination therapy is generally more effective than monotherapy in the field of cancer treatment. However, screening for effective synergistic combinations from a wide range of drug combinations is particularly important given the increase in the number of available drug classes and potential drug-drug interactions. Existing methods for predicting the synergistic effects of drug combinations primarily focus on extracting structural features of drug molecules and cell lines, but neglect the interaction mechanisms between cell lines and drug combinations. Consequently, there is a deficiency in comprehensive understanding of the synergistic effects of drug combinations. To address this issue, we propose a drug combination synergy prediction model based on multi-source feature interaction learning, named MFSynDCP, aiming to predict the synergistic effects of anti-tumor drug combinations. This model includes a graph aggregation module with an adaptive attention mechanism for learning drug interactions and a multi-source feature interaction learning controller for managing information transfer between different data sources, accommodating both drug and cell line features. Comparative studies with benchmark datasets demonstrate MFSynDCP's superiority over existing methods. Additionally, its adaptive attention mechanism graph aggregation module identifies drug chemical substructures crucial to the synergy mechanism. Overall, MFSynDCP is a robust tool for predicting synergistic drug combinations. The source code is available from GitHub at https://github.com/kkioplkg/MFSynDCP .

MMGAT: a graph attention network framework for ATAC-seq motifs finding.

BACKGROUND: Motif finding in Assay for Transposase-Accessible Chromatin using sequencing (ATAC-seq) data is essential to reveal the intricacies of transcription factor binding sites (TFBSs) and their pivotal roles in gene regulation. Deep learning technologies including convolutional neural networks (CNNs) and graph neural networks (GNNs), have achieved success in finding ATAC-seq motifs. However, CNN-based methods are limited by the fixed width of the convolutional kernel, which makes it difficult to find multiple transcription factor binding sites with different lengths. GNN-based methods has the limitation of using the edge weight information directly, makes it difficult to aggregate the neighboring nodes' information more efficiently when representing node embedding. RESULTS: To address this challenge, we developed a novel graph attention network framework named MMGAT, which employs an attention mechanism to adjust the attention coefficients among different nodes. And then MMGAT finds multiple ATAC-seq motifs based on the attention coefficients of sequence nodes and k-mer nodes as well as the coexisting probability of k-mers. Our approach achieved better performance on the human ATAC-seq datasets compared to existing tools, as evidenced the highest scores on the precision, recall, F1_score, ACC, AUC, and PRC metrics, as well as finding 389 higher quality motifs. To validate the performance of MMGAT in predicting TFBSs and finding motifs on more datasets, we enlarged the number of the human ATAC-seq datasets to 180 and newly integrated 80 mouse ATAC-seq datasets for multi-species experimental validation. Specifically on the mouse ATAC-seq dataset, MMGAT also achieved the highest scores on six metrics and found 356 higher-quality motifs. To facilitate researchers in utilizing MMGAT, we have also developed a user-friendly web server named MMGAT-S that hosts the MMGAT method and ATAC-seq motif finding results. CONCLUSIONS: The advanced methodology MMGAT provides a robust tool for finding ATAC-seq motifs, and the comprehensive server MMGAT-S makes a significant contribution to genomics research. The open-source code of MMGAT can be found at https://github.com/xiaotianr/MMGAT , and MMGAT-S is freely available at https://www.mmgraphws.com/MMGAT-S/ .