College of Computer Science and Engineering, Chongqing University of Technology

【目的】阐明药物和基因之间的相互联系是药物开发中的一个重要课题。目前,基于随机游走算法的图神经网络方法在解决药物与基因交互关系识别上已经取得了不错的效果,但是当前的方法,单一子图的方法往往容易忽略掉全局图的信息,不能够很好地聚合节点的信息,同时,药物和基因的节点表示采用简单的融合方式,不能够有效地利用节点表示的信息,用于交互关系的分类。针对上述问题提出了跨粒度对比学习与注意力融合的药物-基因交互关系预测方法。【方法】一方面采用跨粒度的对比学习方法,得到远距离和近距离的节点信息,同时采用对比学习的结构增加对药物和基因节点的区分。另一方面利用注意力融合机制,充分挖掘节点中隐含的信息,将远近距离信息进行注意力融合。【结果】在2个真实数据集上的实验结果表明该模型比基线模型具有更好的分类效果。

【Purposes】Clarifying the interconnections between drugs and genes is an important topic in drug development. At present, the graph neural network method based on the random walk al‐ gorithm has achieved great results in identifying drug-gene interaction relationships. However, exist‐ ing methods with single graph neural network modeling can’t aggregate the information of neighbor nodes well. In addition, most methods use a simple way for the node representation of drugs and genes fusing, which fais to effectively use the information represented by nodes for the classification of interaction relationships. To address the above issues, a cross granularity contrastive learning and at‐ tention fusion method is proposed for predicting drug-gene interaction relationships. 【Methods】On one hand, a cross granularity contrastive learning method is adopted to obtain node information from both distant and close distances. On the other hand, by utilizing attention fusion mechanisms, hidden information in nodes can be fully mined, and attention fusion can be performed on distance informa‐ tion. 【Findings】 The experimental results on two real datasets show that compared with the baseline; the proposed model has better classification performance.

国家自然科学基金项目(62141201);重庆市自然科学基金项目(CSTB2022NSCQ-MSX1672);重庆市教育委员会科学技术研究计划重大项目(KJZD-M202201102)

胡冬冬,彭杨,谭暑秋,等.跨粒度子图对比学习与注意力融合的药物-基因关系预测[J].太原理工大学学报,2025,56(1):127-136.

HU Dongdong,PENG Yang,TAN Shuqiu,et al.Drug-gene interaction prediction method through cross granular‐ity subgraph contrastive learning and attention mechanism fusion[J].Journal of Taiyuan University of Technoloty,2025,56(1):127-136.