HAMDA: Hybrid Approach for MiRNA-Disease Association prediction
HAMDA: Hybrid Approach for MiRNA-Disease Association prediction
复制标题
DOI：
10.1016/j.jbi.2017.10.014
复制
发表时间：
2017-12-01
影响因子：
4.5
通讯作者：
Yan, Gui-Ying
中科院分区：
文献类型：
--
作者：
Chen, Xing;Niu, Ya-Wei;Yan, Gui-Ying
几十年来，大量的实验研究共同表明，microRNA （miRNA）在许多关键的生物学过程中发挥着不可或缺的作用，因此也是人类复杂疾病的发病机制。由于传统生物学实验需要耗费大量的资源和时间，因此开发有效可行的计算方法来预测疾病与miRNA之间的潜在关联越来越受到人们的重视。在这项研究中，我们开发了一个基于混合图推荐算法的miRNA-疾病关联预测混合方法（HAMDA）计算模型，通过将实验验证的miRNA-疾病关联、疾病语义相似度、miRNA功能相似度和高斯相互作用谱核相似度整合到推荐算法中，揭示新的miRNA-疾病关联。HAMDA既考虑了网络结构和信息传播，又考虑了节点属性，预测效果令人满意。HAMDA在全局和局部留一交叉验证框架下的auc分别为0.9035和0.8395。同时，HAMDA在5倍交叉验证中也取得了良好的效果，AUC为0.8965 +/- 0.0012。此外，我们对三种重要的人类癌症进行了案例研究，以评估HAMDA的性能。因此，在最近的实验文献中，预测前50位的mirna中有90%（淋巴瘤）、86%（前列腺癌）和92%（肾癌）被证实，表明HAMDA具有可靠的预测能力。
For decades, enormous experimental researches have collectively indicated that microRNA (miRNA) could play indispensable roles in many critical biological processes and thus also the pathogenesis of human complex diseases. Whereas the resource and time cost required in traditional biology experiments are expensive, more and more attentions have been paid to the development of effective and feasible computational methods for predicting potential associations between disease and miRNA. In this study, we developed a computational model of Hybrid Approach for MiRNA-Disease Association prediction (HAMDA), which involved the hybrid graph-based recommendation algorithm, to reveal novel miRNA-disease associations by integrating experimentally verified miRNA-disease associations, disease semantic similarity, miRNA functional similarity, and Gaussian interaction profile kernel similarity into a recommendation algorithm. HAMDA took not only network structure and information propagation but also node attribution into consideration, resulting in a satisfactory prediction performance. Specifically, HAMDA obtained AUCs of 0.9035 and 0.8395 in the frameworks of global and local leave-one-out cross validation, respectively. Meanwhile, HAMDA also achieved good performance with AUC of 0.8965 +/- 0.0012 in 5-fold cross validation. Additionally, we conducted case studies about three important human cancers for performance evaluation of HAMDA. As a result, 90% (Lymphoma), 86% (Prostate Cancer) and 92% (Kidney Cancer) of top 50 predicted miRNAs were confirmed by recent experiment literature, which showed the reliable prediction ability of HAMDA.