基于多任务学习和超图神经网络的微生物-药物关联预测
- 分享到:
摘要：
传统的生物实验方法寻找微生物与药物关系不仅耗时费力,而且成本极高.因此,为了降低实验成本并提高效率,计算方法被用于预测微生物-药物关联.然而,现有方法忽视了疾病作为中介的关键作用,导致数据稀疏性问题.为此,提出了基于多任务学习的模型(MTLTPMDA),用于同时预测微生物-药物和疾病-药物关联.模型通过共享药物节点的特征来增强任务间的联系,并利用超图神经网络(HGNN)探索微生物、药物和疾病之间的复杂交互.通过构建微生物-药物和疾病-药物超图,HGNN有效捕捉了多节点间的高阶关系,在五重交叉验证下,MTLTPMDA实现了AUC为0.9033和AUPR为0.8930,优于多种现有方法,展示了模型在预测潜在关联上的有效性.
Traditional biological experiments to discover microbe-drug relationships are not only time-consuming and Ia-bor-intensive but also highy expensive Therefore, to redtce expertmnental costs and improve effrdteney, computattonal methods have been employed to predict microbe-drug associations. However, the existing methods neglect the crucial role of diseases as intermediaries, which leads to the problem of data sparsity. To address this, we propose a multi-task learning model(MTLT-PMDA) that simultaneously predicts microbe-drug and disease-drug associations. The model enhances the connections between tasks by sharing drug node features and utilizes a hypergraph neural network(HGNN) to explore the complex interactions between microbes, drugs, and diseases. By constructing microbe-drug and disease-drug hypergraphs, the HGNN effectively cap-tures Highe-order relaronships amnong mutnple nodes. In a five-fold cross-valridafon framnework , MTLTPMDA adhieved an AUC of 0.9033 and an AUPR of O.8930, outperforming several existing methods, demonstrating the model's effectiveness in predicting potential associations.
作者:
王波，王钧祺，杜晓昕，孙明，王彤轩，黎景威
Wang Bo,Wang Junqi,Du Xiaoxin,Sun Ming,Wang Tongxuan,Li Jingwei
机构地区：
齐齐哈尔大学计算机与控制工程学院；黑龙江省大数据网络安全检测分析重点实验室
引用本文：
王波，王钧祺，杜晓昕等。基于多任务学习和超图神经网络的微生物-药物关联预测 [J].河南师范大学学报(自然科学版),2026,54(1):68-76.(Wang Bo, Wang Jungi, Du Xiaoxin,et al.Predicting microbe-drug associations based on multi-task learning and hypergraph neural network[J].Journal of Henan Normal University(Natural Science Edition),2026,54(1):68-76.DOI:10.16366/j.cnki.1000-2367.2024.10.08.0001.)
基金：
黑龙江省省属高等学校基本科研业务费国自然培育一般项目
关键词：
微生物与药物关联；疾病与药物关联；多任务学习技术；数据稀疏性；超图神经网络
microbe-drug associations; disease-drug associations; muli-task learning technology; data sparsity; hyperg-raph neural network
分类号：
TP391
