Computational approaches for predicting drug-target interactions(DTIs)are pivotal in advancing drug discovery.Current methodologies leveraging heterogeneous networks often fall short in fully integrating both local an...Computational approaches for predicting drug-target interactions(DTIs)are pivotal in advancing drug discovery.Current methodologies leveraging heterogeneous networks often fall short in fully integrating both local and global network information.To comprehensively consider network information,we propose DHGT-DTI,a novel deep learning-based approach for DTI prediction.Specifically,we capture the local and global structural information of the network from both neighborhood and meta-path per-spectives.In the neighborhood perspective,we employ a heterogeneous graph neural network(HGNN),which extends Graph Sample and Aggregate(GraphSAGE)to handle diverse node and edge types,effectively learning local network structures.In the meta-path perspective,we introduce a Graph Transformer with residual connections to model higher-order relationships defined by meta-paths,such as"drug-disease-drug",and use an attention mechanism to fuse information across multiple meta-paths.The learned features from these dual perspectives are synergistically integrated for DTI prediction via a matrix decomposition method.Furthermore,DHGT-DTI reconstructs not only the DTI network but also auxiliary networks to bolster prediction accuracy.Comprehensive experiments on two benchmark datasets validate the superiority of DHGT-DTI over existing baseline methods.Additionally,case studies on six drugs used to treat Parkinson's disease not only validate the practical utility of DHGT-DTI but also highlight its broader potential in accelerating drug discovery for other diseases.展开更多
Graph convolutional networks(GCNs)have received significant attention from various research fields due to the excellent performance in learning graph representations.Although GCN performs well compared with other meth...Graph convolutional networks(GCNs)have received significant attention from various research fields due to the excellent performance in learning graph representations.Although GCN performs well compared with other methods,it still faces challenges.Training a GCN model for large-scale graphs in a conventional way requires high computation and storage costs.Therefore,motivated by an urgent need in terms of efficiency and scalability in training GCN,sampling methods have been proposed and achieved a significant effect.In this paper,we categorize sampling methods based on the sampling mechanisms and provide a comprehensive survey of sampling methods for efficient training of GCN.To highlight the characteristics and differences of sampling methods,we present a detailed comparison within each category and further give an overall comparative analysis for the sampling methods in all categories.Finally,we discuss some challenges and future research directions of the sampling methods.展开更多
基于深度学习的网络攻击检测是对欧几里得数据进行建模,无法学习攻击数据中的结构特征。为此,提出一种基于改进图采样与聚合(graph sample and aggregate,GraphSAGE)的网络攻击检测算法。首先,将攻击数据从平面结构转换为图结构数据。其...基于深度学习的网络攻击检测是对欧几里得数据进行建模,无法学习攻击数据中的结构特征。为此,提出一种基于改进图采样与聚合(graph sample and aggregate,GraphSAGE)的网络攻击检测算法。首先,将攻击数据从平面结构转换为图结构数据。其次,对GraphSAGE算法进行了改进,包括在消息传递阶段融合节点和边的特征,同时在消息聚合过程中考虑不同源节点对目标节点的影响程度,并在边嵌入生成时引入残差学习机制。在两个公开网络攻击数据集上的实验结果表明,在二分类情况下,所提算法的总体性能优于E-GraphSAGE、LSTM、RNN、CNN算法;在多分类情况下,所提算法在大多数攻击类型上的F1值高于对比算法。展开更多
Graph Neural Networks(GNNs),as a deep learning framework specifically designed for graph-structured data,have achieved deep representation learning of graph data through message passing mechanisms and have become a co...Graph Neural Networks(GNNs),as a deep learning framework specifically designed for graph-structured data,have achieved deep representation learning of graph data through message passing mechanisms and have become a core technology in the field of graph analysis.However,current reviews on GNN models are mainly focused on smaller domains,and there is a lack of systematic reviews on the classification and applications of GNN models.This review systematically synthesizes the three canonical branches of GNN,Graph Convolutional Network(GCN),Graph Attention Network(GAT),and Graph Sampling Aggregation Network(GraphSAGE),and analyzes their integration pathways from both structural and feature perspectives.Drawing on representative studies,we identify three major integration patterns:cascaded fusion,where heterogeneous modules such as Convolutional Neural Network(CNN),Long Short-Term Memory(LSTM),and GraphSAGE are sequentially combined for hierarchical feature learning;parallel fusion,where multi-branch architectures jointly encode complementary graph features;and feature-level fusion,which employs concatenation,weighted summation,or attention-based gating to adaptively merge multi-source embeddings.Through these patterns,integrated GNNs achieve enhanced expressiveness,robustness,and scalability across domains including transportation,biomedicine,and cybersecurity.展开更多
目的针对电子鼻在温湿度波动较大的医疗环境和户外等场景中因传感器漂移导致检测失效的问题,提出小样本补偿模型,解决传统方法依赖大量漂移数据、难以适应长期非线性漂移的瓶颈。方法构建传感器漂移适应中的图神经网络(graph neural net...目的针对电子鼻在温湿度波动较大的医疗环境和户外等场景中因传感器漂移导致检测失效的问题,提出小样本补偿模型,解决传统方法依赖大量漂移数据、难以适应长期非线性漂移的瓶颈。方法构建传感器漂移适应中的图神经网络(graph neural network used in sensors drift adaptation,GNNSD)模型,融合深度残差卷积与图神经网络,采用数据增强与关系推理机制,在公开传感器漂移数据集上开展小样本分类实验。结果GNNSD模型在K=1设置下实现84.12%平均准确率,较最优对比算法FEDA提升9.93%。消融实验表明模型架构具有合理性。结论该模型通过多尺度特征与图结构关系推理的协同机制,当每个类别的参考样本数量只有1个时也可实现较高分类精度,为医疗监测、跨境筛查等生物安全场景提供低样本依赖的漂移补偿解决方案。展开更多
针对群推荐系统的数据稀疏性挑战,以及现有群推荐方法忽略群、群成员以及不同备选物品之间的复杂关联关系问题,提出一种基于图注意力网络的群推荐方法(group recommendation method based on graph attention network,GAT-GRM)。首先,...针对群推荐系统的数据稀疏性挑战,以及现有群推荐方法忽略群、群成员以及不同备选物品之间的复杂关联关系问题,提出一种基于图注意力网络的群推荐方法(group recommendation method based on graph attention network,GAT-GRM)。首先,将群推荐系统中群、群成员以及不同物品间的复杂关联刻画为层次关系图数据,包括用户-物品交互关系图、群组-用户包含关系图、群组-物品交互关系图等。其次,利用图注意力网络内在地聚合各类交互关系图,从历史交互数据中动态学习群偏好、用户偏好和物品特征。最后,基于群偏好、用户偏好和物品特征进行物品评分预测。实验结果表明,在CAMRa2011数据集上,GAT-GRM的性能显著优于各类基准算法。在稀疏度98.89%的群推荐任务下,GAT-GRM平均绝对偏差和均方根误差相较最优基准算法分别降低9.3%和9.6%。展开更多
Supramolecular assembly is a versatile bottom-up strategy for creating advanced functional materials.Metallic platinum–platinum(Pt…Pt)interactions provide a distinctive driving force for supramolecular assembly due ...Supramolecular assembly is a versatile bottom-up strategy for creating advanced functional materials.Metallic platinum–platinum(Pt…Pt)interactions provide a distinctive driving force for supramolecular assembly due to their strong,directional,and longrange nature.Despite their importance,the microscopic dynamics underlying the self-assembly of Pt(II)complexes remain challenging to probe experimentally.Molecular dynamics(MD)simulations can capture these processes at atomic resolution,but extracting kinetic pathways is complicated by the indistinguishability and permutation of identical monomers within selfassembled structures.In this study,we employ GraphVAMPnet,a deep learning framework based on graph neural networks(GNN),on extensive MD simulations of amphiphilic PtB complexes during the early stage of self-assembly.GraphVAMPnet inherently accounts for permutational,rotational,and translational invariance,making it well-suited for analyzing self-assembly dynamics.Our analysis reveals three slow collective variables(CVs)that govern PtB self-assembly.The slowest mode(CV1)separates two distinct kinetic growth routes:an incremental growth mechanism,in which single monomers join existing aggregates with predominantly antiparallel packing between two adjacent PtB complexes(CV3),and a hopping growth mechanism,in which clusters of smaller size merge via heterogeneous collisions,yielding a mix of antiparallel and parallel packing arrangements(CV2).Further energetic analysis indicates that incremental growth is favored,potentially leading to the well-ordered nanosheet morphologies observed experimentally.Our findings provide molecular-level insight into PtB selfassembly pathways and showcase the capability of GraphVAMPnet in dissecting the complex dynamics of supramolecular assembly.展开更多
Identifying cancer driver genes has paramount significance in elucidating the intricate mechanisms underlying cancer development,progression,and therapeutic interventions.Abundant omics data and interactome networks p...Identifying cancer driver genes has paramount significance in elucidating the intricate mechanisms underlying cancer development,progression,and therapeutic interventions.Abundant omics data and interactome networks provided by numerous extensive databases enable the application of graph deep learning techniques that incorporate network structures into the deep learning framework.However,most existing models primarily focus on individual network,inevitably neglecting the incompleteness and noise of interactions.Moreover,samples with imbalanced classes in driver gene identification hamper the performance of models.To address this,we propose a novel deep learning framework MMGN,which integrates multiplex networks and pan-cancer multiomics data using graph neural networks combined with negative sample inference to discover cancer driver genes,which not only enhances gene feature learning based on the mutual information and the consensus regularizer,but also achieves balanced class of positive and negative samples for model training.The reliability of MMGN has been verified by the Area Under the Receiver Operating Characteristic curves(AUROC)and the Area Under the Precision-Recall Curves(AUPRC).We believe MMGN has the potential to provide new prospects in precision oncology and may find broader applications in predicting biomarkers for other intricate diseases.展开更多
阿尔茨海默病(Alzheimer’s Disease,AD)是一种慢性神经系统退行性疾病,其准确分类有助于实现AD的早期诊断,从而及时采取针对性的治疗和干预措施.本文提出了一种最近邻域聚合图神经网络(Graph neural network with nearest Neighborhood...阿尔茨海默病(Alzheimer’s Disease,AD)是一种慢性神经系统退行性疾病,其准确分类有助于实现AD的早期诊断,从而及时采取针对性的治疗和干预措施.本文提出了一种最近邻域聚合图神经网络(Graph neural network with nearest Neighborhood AgGrEgation,GraphNAGE)的AD分类新方法.首先进行图数据建模,将AD数据样本表示为图数据.采用基于互信息(Mutual Information,MI)的特征选择方法,从样本的114维大脑皮层与皮层下感兴趣区域(Cerebral Cortex and Subcortical Regions Of Interest,CCS-ROI)的体积特征中选取重要性高的体积特征,并将其用于节点建模.提出基于相似性度量的关系建模方法,利用重要性高的体积特征、遗传基因、人口统计信息和认知评分对样本之间的关系进行建模.进而构建GraphNAGE,针对每个节点,基于与该节点相关的边的权重进行最近邻域采样,然后使用均值聚合方法对采样得到的邻居节点和中心节点的数据进行聚合,最后通过一个全连接层和一个Softmax层实现AD分类.在TADPOLE(The Alzheimer’s Disease Prediction Of Longitudinal Evolution)数据集上进行实验,结果表明:本文提出的AD分类方法的准确率(ACCuracy,ACC)为98.20%,F_(1)分数为97.34%,曲线下面积(Area Under Curve,AUC)为97.80%.实验结果表明:本文提出的AD分类方法充分利用了AD数据样本之间的相关性,其性能优于传统的基于机器学习、深度学习和图神经网络(Graph Neural Network,GNN)的AD分类方法.展开更多
基金the National Natural Science Foundation of China(Grant Nos.:62272288,U22A2041)Fundamental Research Funds for the Central Universities,Shaanxi Normal University(Grant No.:GK202302006)the Scientific Research Fund of Hunan Provincial Education Department of China(Grant No.:22B0097).
文摘Computational approaches for predicting drug-target interactions(DTIs)are pivotal in advancing drug discovery.Current methodologies leveraging heterogeneous networks often fall short in fully integrating both local and global network information.To comprehensively consider network information,we propose DHGT-DTI,a novel deep learning-based approach for DTI prediction.Specifically,we capture the local and global structural information of the network from both neighborhood and meta-path per-spectives.In the neighborhood perspective,we employ a heterogeneous graph neural network(HGNN),which extends Graph Sample and Aggregate(GraphSAGE)to handle diverse node and edge types,effectively learning local network structures.In the meta-path perspective,we introduce a Graph Transformer with residual connections to model higher-order relationships defined by meta-paths,such as"drug-disease-drug",and use an attention mechanism to fuse information across multiple meta-paths.The learned features from these dual perspectives are synergistically integrated for DTI prediction via a matrix decomposition method.Furthermore,DHGT-DTI reconstructs not only the DTI network but also auxiliary networks to bolster prediction accuracy.Comprehensive experiments on two benchmark datasets validate the superiority of DHGT-DTI over existing baseline methods.Additionally,case studies on six drugs used to treat Parkinson's disease not only validate the practical utility of DHGT-DTI but also highlight its broader potential in accelerating drug discovery for other diseases.
基金supported by the National Natural Science Foundation of China(61732018,61872335,61802367,61876215)the Strategic Priority Research Program of Chinese Academy of Sciences(XDC05000000)+1 种基金Beijing Academy of Artificial Intelligence(BAAI),the Open Project Program of the State Key Laboratory of Mathematical Engineering and Advanced Computing(2019A07)the Open Project of Zhejiang Laboratory,and a grant from the Institute for Guo Qiang,Tsinghua University.Recommended by Associate Editor Long Chen.
文摘Graph convolutional networks(GCNs)have received significant attention from various research fields due to the excellent performance in learning graph representations.Although GCN performs well compared with other methods,it still faces challenges.Training a GCN model for large-scale graphs in a conventional way requires high computation and storage costs.Therefore,motivated by an urgent need in terms of efficiency and scalability in training GCN,sampling methods have been proposed and achieved a significant effect.In this paper,we categorize sampling methods based on the sampling mechanisms and provide a comprehensive survey of sampling methods for efficient training of GCN.To highlight the characteristics and differences of sampling methods,we present a detailed comparison within each category and further give an overall comparative analysis for the sampling methods in all categories.Finally,we discuss some challenges and future research directions of the sampling methods.
文摘基于深度学习的网络攻击检测是对欧几里得数据进行建模,无法学习攻击数据中的结构特征。为此,提出一种基于改进图采样与聚合(graph sample and aggregate,GraphSAGE)的网络攻击检测算法。首先,将攻击数据从平面结构转换为图结构数据。其次,对GraphSAGE算法进行了改进,包括在消息传递阶段融合节点和边的特征,同时在消息聚合过程中考虑不同源节点对目标节点的影响程度,并在边嵌入生成时引入残差学习机制。在两个公开网络攻击数据集上的实验结果表明,在二分类情况下,所提算法的总体性能优于E-GraphSAGE、LSTM、RNN、CNN算法;在多分类情况下,所提算法在大多数攻击类型上的F1值高于对比算法。
基金funded by Guangzhou Huashang University(2024HSZD01,HS2023JYSZH01).
文摘Graph Neural Networks(GNNs),as a deep learning framework specifically designed for graph-structured data,have achieved deep representation learning of graph data through message passing mechanisms and have become a core technology in the field of graph analysis.However,current reviews on GNN models are mainly focused on smaller domains,and there is a lack of systematic reviews on the classification and applications of GNN models.This review systematically synthesizes the three canonical branches of GNN,Graph Convolutional Network(GCN),Graph Attention Network(GAT),and Graph Sampling Aggregation Network(GraphSAGE),and analyzes their integration pathways from both structural and feature perspectives.Drawing on representative studies,we identify three major integration patterns:cascaded fusion,where heterogeneous modules such as Convolutional Neural Network(CNN),Long Short-Term Memory(LSTM),and GraphSAGE are sequentially combined for hierarchical feature learning;parallel fusion,where multi-branch architectures jointly encode complementary graph features;and feature-level fusion,which employs concatenation,weighted summation,or attention-based gating to adaptively merge multi-source embeddings.Through these patterns,integrated GNNs achieve enhanced expressiveness,robustness,and scalability across domains including transportation,biomedicine,and cybersecurity.
文摘目的针对电子鼻在温湿度波动较大的医疗环境和户外等场景中因传感器漂移导致检测失效的问题,提出小样本补偿模型,解决传统方法依赖大量漂移数据、难以适应长期非线性漂移的瓶颈。方法构建传感器漂移适应中的图神经网络(graph neural network used in sensors drift adaptation,GNNSD)模型,融合深度残差卷积与图神经网络,采用数据增强与关系推理机制,在公开传感器漂移数据集上开展小样本分类实验。结果GNNSD模型在K=1设置下实现84.12%平均准确率,较最优对比算法FEDA提升9.93%。消融实验表明模型架构具有合理性。结论该模型通过多尺度特征与图结构关系推理的协同机制,当每个类别的参考样本数量只有1个时也可实现较高分类精度,为医疗监测、跨境筛查等生物安全场景提供低样本依赖的漂移补偿解决方案。
文摘针对群推荐系统的数据稀疏性挑战,以及现有群推荐方法忽略群、群成员以及不同备选物品之间的复杂关联关系问题,提出一种基于图注意力网络的群推荐方法(group recommendation method based on graph attention network,GAT-GRM)。首先,将群推荐系统中群、群成员以及不同物品间的复杂关联刻画为层次关系图数据,包括用户-物品交互关系图、群组-用户包含关系图、群组-物品交互关系图等。其次,利用图注意力网络内在地聚合各类交互关系图,从历史交互数据中动态学习群偏好、用户偏好和物品特征。最后,基于群偏好、用户偏好和物品特征进行物品评分预测。实验结果表明,在CAMRa2011数据集上,GAT-GRM的性能显著优于各类基准算法。在稀疏度98.89%的群推荐任务下,GAT-GRM平均绝对偏差和均方根误差相较最优基准算法分别降低9.3%和9.6%。
基金funding from the Wisconsin Alumni Research Foundation.
文摘Supramolecular assembly is a versatile bottom-up strategy for creating advanced functional materials.Metallic platinum–platinum(Pt…Pt)interactions provide a distinctive driving force for supramolecular assembly due to their strong,directional,and longrange nature.Despite their importance,the microscopic dynamics underlying the self-assembly of Pt(II)complexes remain challenging to probe experimentally.Molecular dynamics(MD)simulations can capture these processes at atomic resolution,but extracting kinetic pathways is complicated by the indistinguishability and permutation of identical monomers within selfassembled structures.In this study,we employ GraphVAMPnet,a deep learning framework based on graph neural networks(GNN),on extensive MD simulations of amphiphilic PtB complexes during the early stage of self-assembly.GraphVAMPnet inherently accounts for permutational,rotational,and translational invariance,making it well-suited for analyzing self-assembly dynamics.Our analysis reveals three slow collective variables(CVs)that govern PtB self-assembly.The slowest mode(CV1)separates two distinct kinetic growth routes:an incremental growth mechanism,in which single monomers join existing aggregates with predominantly antiparallel packing between two adjacent PtB complexes(CV3),and a hopping growth mechanism,in which clusters of smaller size merge via heterogeneous collisions,yielding a mix of antiparallel and parallel packing arrangements(CV2).Further energetic analysis indicates that incremental growth is favored,potentially leading to the well-ordered nanosheet morphologies observed experimentally.Our findings provide molecular-level insight into PtB selfassembly pathways and showcase the capability of GraphVAMPnet in dissecting the complex dynamics of supramolecular assembly.
基金supported in part by the National Natural Science Foundation of China(No.62202383)the Guangdong Basic and Applied Basic Research Foundation(No.2024A1515012602)the National Key Research and Development Program of China(No.2022YFD1801200).
文摘Identifying cancer driver genes has paramount significance in elucidating the intricate mechanisms underlying cancer development,progression,and therapeutic interventions.Abundant omics data and interactome networks provided by numerous extensive databases enable the application of graph deep learning techniques that incorporate network structures into the deep learning framework.However,most existing models primarily focus on individual network,inevitably neglecting the incompleteness and noise of interactions.Moreover,samples with imbalanced classes in driver gene identification hamper the performance of models.To address this,we propose a novel deep learning framework MMGN,which integrates multiplex networks and pan-cancer multiomics data using graph neural networks combined with negative sample inference to discover cancer driver genes,which not only enhances gene feature learning based on the mutual information and the consensus regularizer,but also achieves balanced class of positive and negative samples for model training.The reliability of MMGN has been verified by the Area Under the Receiver Operating Characteristic curves(AUROC)and the Area Under the Precision-Recall Curves(AUPRC).We believe MMGN has the potential to provide new prospects in precision oncology and may find broader applications in predicting biomarkers for other intricate diseases.
文摘阿尔茨海默病(Alzheimer’s Disease,AD)是一种慢性神经系统退行性疾病,其准确分类有助于实现AD的早期诊断,从而及时采取针对性的治疗和干预措施.本文提出了一种最近邻域聚合图神经网络(Graph neural network with nearest Neighborhood AgGrEgation,GraphNAGE)的AD分类新方法.首先进行图数据建模,将AD数据样本表示为图数据.采用基于互信息(Mutual Information,MI)的特征选择方法,从样本的114维大脑皮层与皮层下感兴趣区域(Cerebral Cortex and Subcortical Regions Of Interest,CCS-ROI)的体积特征中选取重要性高的体积特征,并将其用于节点建模.提出基于相似性度量的关系建模方法,利用重要性高的体积特征、遗传基因、人口统计信息和认知评分对样本之间的关系进行建模.进而构建GraphNAGE,针对每个节点,基于与该节点相关的边的权重进行最近邻域采样,然后使用均值聚合方法对采样得到的邻居节点和中心节点的数据进行聚合,最后通过一个全连接层和一个Softmax层实现AD分类.在TADPOLE(The Alzheimer’s Disease Prediction Of Longitudinal Evolution)数据集上进行实验,结果表明:本文提出的AD分类方法的准确率(ACCuracy,ACC)为98.20%,F_(1)分数为97.34%,曲线下面积(Area Under Curve,AUC)为97.80%.实验结果表明:本文提出的AD分类方法充分利用了AD数据样本之间的相关性,其性能优于传统的基于机器学习、深度学习和图神经网络(Graph Neural Network,GNN)的AD分类方法.
