Scholarlink

<p>Adverse Drug Reactions (ADRs) pose significant challenges in clinical practice, particularly in oncology, where treatment regimens are complex and patient responses are highly variable. Researchers have developed various statistical based, neural network based and graph-based algorithms for biomedical tasks. Existing neural network-based methods for ADR prediction and knowledge discovery suffer from several limitations: they often treat all neighbors equally in Graph Neural Network (GNN) models, and fail to capture complex, intersentential, high granular, indirect information from neighboring nodes. Additionally, these models do not capture causal dependencies, a crucial factor in the biomedical domain. To address these limitations, we propose the Causality and Proximity-based Relational Multihead Attention Model (CPRMAM). This model leverages a knowledge graph representing ADR-related case studies of cancer patients, extracting and concatenating node features to obtain comprehensive feature representations and propose a novel aggregation function based on causal proximity vector.</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

Enhanced GNN with causal proximity vectors: a neural network based methodology for predicting adverse drug reaction in cancer

  • 期刊论文

  • 发表日期：

    2026年1月7日

查看全文
  • Samridhi Dev,
  • Aditi Sharan

摘要

Adverse Drug Reactions (ADRs) pose significant challenges in clinical practice, particularly in oncology, where treatment regimens are complex and patient responses are highly variable. Researchers have developed various statistical based, neural network based and graph-based algorithms for biomedical tasks. Existing neural network-based methods for ADR prediction and knowledge discovery suffer from several limitations: they often treat all neighbors equally in Graph Neural Network (GNN) models, and fail to capture complex, intersentential, high granular, indirect information from neighboring nodes. Additionally, these models do not capture causal dependencies, a crucial factor in the biomedical domain. To address these limitations, we propose the Causality and Proximity-based Relational Multihead Attention Model (CPRMAM). This model leverages a knowledge graph representing ADR-related case studies of cancer patients, extracting and concatenating node features to obtain comprehensive feature representations and propose a novel aggregation function based on causal proximity vector.