基于图神经网络的环状RNA生物标志物筛选预测算法  

Prediction algorithm for screening circRNA biomarker based on graph neural network

作　　者：李扬[1] 胡学钢[1] 王磊[2] 李培培[1] 尤著宏 Yang LI;Xuegang HU;Lei WANG;Peipei LI;Zhuhong YOU(School of Computer Science and Information Engineering,Hefei University of Technology,Hefei 230601,China;College of Information Science and Engineering,Zaozhuang University,Zaozhuang 277160,China;School of Computer Science,Northwestern Polytechnical University,Xi'an 710129,China)

机构地区：[1]合肥工业大学计算机与信息学院,合肥230601 [2]枣庄学院信息科学与工程学院,枣庄277160 [3]西北工业大学计算机学院,西安710129

出　　处：《中国科学：信息科学》2023年第11期2214-2229,共16页Scientia Sinica(Informationis)

基　　金：国家自然科学基金(批准号:62076085,62172355,61702444);国家重点研发计划(批准号:2016YFB1000901);中央高校基本科研业务费专项资金(批准号:JZ2020HGQA0186)资助项目。

摘　　要：越来越多的证据表明,环状RNA(circular RNA,circRNA)在人类复杂疾病发病机制和许多重要生物学过程中发挥不可或缺的作用.确定环状RNA与疾病之间关联对于复杂人类疾病的诊断和治疗具有重要的潜在价值.然而,传统的湿实验方式通常是盲目、低效、耗时且昂贵的,往往还伴随着高的假阳性率.因此,迫切需要有效和可行的计算方法来大规模预测潜在的环状RNA–疾病关联.本文通过结合图神经网络的高阶图卷积网络算法与随机蕨分类器对环状RNA与疾病之间的关联关系进行预测.该方法能够从环状RNA和疾病多种属性信息构建的多源相似性网络中,有效抽取具有高阶混合邻域信息的高级特征,并对其进行准确分类.在5折交叉验证实验中,该方法在CircR2Disease数据集上取得了93.75%的AUC得分.此外,在案例研究中,该模型的预测结果得到了生物湿实验的支持,预测得分前15的环状RNA–疾病关联中的13个在最近发表文献中得以证实.这些优异的结果表明,所提模型是预测环状RNA–疾病关联的有效工具,并且可以为生物湿实验提供理论依据和高可信的环状RNA候选生物标志物.Emerging evidence has revealed that circular RNA(circRNA)plays an indispensable role in the pathogenesis of complex human diseases and various biological processes.Identifying the associations between circRNAs and diseases is crucial for diagnosing and treating these diseases.However,traditional wet-lab methods are often inefficient,time-consuming,and expensive,with high false-positive rates.Therefore,there is an urgent need for efficient and feasible computational methods to predict potential circRNA-disease associations on a large scale.In this paper,we propose a novel approach to predict the association between circRNA and disease by combining the high-order graph convolutional network algorithm of graph neural network with a random ferns classifier.This approach can effectively extract high-level features with high-order mixed neighborhood information from the multi-source similarity network constructed by multiple attribute information of circRNAs and diseases and accurately classify them.In a 5-fold cross-validation experiment,our method achieved an average AUC score of 93.75%on the CircR2Disease dataset.Furthermore,in case studies,the prediction results of the model were supported by biological wet experiments,and 13 of the top 15 predicted circRNA-disease associations were confirmed by recently published literature.These excellent results indicate that the proposed model is an effective tool for predicting circRNA-disease associations,and can provide a theoretical basis and highly reliable candidate biomarkers of circRNAs for biological wet experiments.

关 键 词：环状RNA 图神经网络 环状RNA–疾病关联 高阶图卷积网络 随机蕨 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] Q811.4[自动化与计算机技术—控制科学与工程]