PDF(Pubmed)
Abstract：
In scenarios involving the treatment of complex or coexisting diseases with multiple drugs, the potential for severe adverse drug reactions in patients necessitates the identification of potential drug-drug interactions (DDIs). Most existing computational methods have not taken into account the asymmetry and relation types of drug interactions caused by the relation information between drugs, which may lead to missing information in embedded learning. Therefore, this paper proposes a directed relation graph attention aware network (DRGATAN) to predict asymmetric drug interactions. DRGATAN leverages an encoder to learn multi-relational role embeddings of drugs across different types of relations. The experimental results show that DRGATAN\'s performance is superior to recognized advanced methods. The visualization demonstrates the effect of utilizing asymmetric information, and the case analysis validates the reliability of the proposed method. This study provides guidance for predicting asymmetric drug interactions.
摘要：
在涉及用多种药物治疗复杂或共存疾病的情况下，患者可能出现严重的药物不良反应,因此需要鉴定潜在的药物-药物相互作用(DDI).现有的大多数计算方法没有考虑药物之间的关系信息引起的药物相互作用的不对称性和关系类型,这可能会导致嵌入式学习中的信息缺失。因此,本文提出了一种有向关系图注意力感知网络（DRGATAN）来预测不对称药物相互作用。DRGATAN利用编码器来学习药物跨不同类型关系的多关系角色嵌入。实验结果表明，DRGATAN的性能优于公认的先进方法。可视化展示了利用不对称信息的效果，实例分析验证了该方法的可靠性。本研究为预测不对称药物相互作用提供了指导。
