Scholarlink

<p>In recent years, graph neural networks (GNNs) have emerged as powerful deep learning (DL) architectures capable of modeling complex relationships in graph-structured data. GNNs have been popularly applied in diverse application domains, including natural language processing, computer vision, information network analysis, as well as bioinformatics. Among these, drug–disease association prediction (DDAP) has become a critical area in bioinformatics, as it plays an essential role in accelerating drug discovery and reducing the cost and time associated with traditional wet-lab experiments. GNNs have demonstrated strong performance in this domain by learning effective representations from graph-modeled biomedical networks, often outperforming traditional machine learning (ML) and DL methods. Despite their success, GNN-based models still face notable challenges. In particular, feature noise as well as task-specific uncertainty often arise due to the complexity and heterogeneity of biomedical data—especially when using multi-layered GNNs. These issues can degrade the quality of learned node representations as well as limit the predictive capability of the model. To address this limitation, we propose a novel model, FG4DDP—a fuzzy-enhanced graph neural network specifically designed for drug–disease association prediction. Our FG4DDP model integrates a fuzzy membership-based neural layer into a graph convolutional network (GCN) architecture. This fuzzy layer serves as a layer-wise uncertainty filtering mechanism, transforming crisp intermediate embeddings into fuzzy degree-based representations that are more resilient to noise and ambiguity. We model the drug–disease associations and related similarity information as a heterogeneous information network (HIN), which serves as the input structure for the FG4DDP learning pipeline. Extensive experiments conducted on a widely-used benchmark DDAP dataset demonstrate that our FG4DDP model consistently outperforms several state-of-the-art ML, DL, and GNN-based models. These results validate the effectiveness of integrating fuzzy logic with GNNs and highlight FG4DDP’s potential as a robust and accurate solution for real-world biomedical association forecasting tasks.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A novel fuzzy-enhanced graph neural network with attention for drug-disease association forecasting problem

  • 期刊论文

  • 发表日期：

    2026年4月15日

查看全文
  • Phu Pham,
  • Giang Tran-Hoang,
  • Trung Nguyen-Huu

摘要

In recent years, graph neural networks (GNNs) have emerged as powerful deep learning (DL) architectures capable of modeling complex relationships in graph-structured data. GNNs have been popularly applied in diverse application domains, including natural language processing, computer vision, information network analysis, as well as bioinformatics. Among these, drug–disease association prediction (DDAP) has become a critical area in bioinformatics, as it plays an essential role in accelerating drug discovery and reducing the cost and time associated with traditional wet-lab experiments. GNNs have demonstrated strong performance in this domain by learning effective representations from graph-modeled biomedical networks, often outperforming traditional machine learning (ML) and DL methods. Despite their success, GNN-based models still face notable challenges. In particular, feature noise as well as task-specific uncertainty often arise due to the complexity and heterogeneity of biomedical data—especially when using multi-layered GNNs. These issues can degrade the quality of learned node representations as well as limit the predictive capability of the model. To address this limitation, we propose a novel model, FG4DDP—a fuzzy-enhanced graph neural network specifically designed for drug–disease association prediction. Our FG4DDP model integrates a fuzzy membership-based neural layer into a graph convolutional network (GCN) architecture. This fuzzy layer serves as a layer-wise uncertainty filtering mechanism, transforming crisp intermediate embeddings into fuzzy degree-based representations that are more resilient to noise and ambiguity. We model the drug–disease associations and related similarity information as a heterogeneous information network (HIN), which serves as the input structure for the FG4DDP learning pipeline. Extensive experiments conducted on a widely-used benchmark DDAP dataset demonstrate that our FG4DDP model consistently outperforms several state-of-the-art ML, DL, and GNN-based models. These results validate the effectiveness of integrating fuzzy logic with GNNs and highlight FG4DDP’s potential as a robust and accurate solution for real-world biomedical association forecasting tasks.