Metapath-Based Deep Convolutional Neural Network for Predicting miRNA-Target Association on Heterogeneous Network,Interdisciplinary Sciences: Computational Life Sciences

当前位置： X-MOL 学术 › Interdiscip. Sci. Comput. Life Sci. › 论文详情

Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)

Metapath-Based Deep Convolutional Neural Network for Predicting miRNA-Target Association on Heterogeneous Network
Interdisciplinary Sciences: Computational Life Sciences ( IF 3.9 ) Pub Date : 2021-06-25 , DOI: 10.1007/s12539-021-00454-3
Jiawei Luo ₁ , Yaoting Bao ₁ , Xiangtao Chen ₁ , Cong Shen ₁

Affiliation

Predicting the interactions between microRNAs (miRNAs) and target genes is of great significance for understanding the regulatory mechanism of miRNA and treating complex diseases. The emergence of large-scale, heterogeneous biological networks has offered unprecedented opportunities for revealing miRNA-associated target genes. However, there are still some limitations about automatically learn the feature information of the network in the existing methods. Since network representation learning can self-adaptively capture structure information of the network, we propose a framework based on heterogeneous network representation, MDCNN (Metapath-Based Deep Convolutional Neural Network), to predict the associations between miRNAs and target genes. MDCNN samples the paths between the node pairs in the form of meta-path based on the heterogeneous information network (HIN) about miRNAs and target genes. Then the node feature and the path feature which is learned by the Deep Convolutional Neural Network (DCNN) are spliced together as the representation of the miRNA-target gene, to predict the miRNA-target gene interactions. The experiment results indicate that the performance of MDCNN outperforms other methods in multiple validation metrics by fivefold cross validation. We set an ablation study to identify the necessity of miRNA similarity and target gene similarity for improving the prediction ability of MDCNN. The case studies on hsa-miR-26b-5p and CDKN1A further demonstrates that MDCNN can successfully predict potential miRNA-target gene interactions.

Graphic abstract

中文翻译：

基于元路径的深度卷积神经网络在异构网络上预测 miRNA-目标关联

预测微小RNA（miRNA）与靶基因之间的相互作用对于理解miRNA的调控机制和治疗复杂疾病具有重要意义。大规模、异质生物网络的出现为揭示 miRNA 相关的靶基因提供了前所未有的机会。然而，现有方法在自动学习网络的特征信息方面还存在一些局限性。由于网络表示学习可以自适应地捕获网络的结构信息，我们提出了一个基于异构网络表示的框架 MDCNN（基于元路径的深度卷积神经网络）来预测 miRNA 与目标基因之间的关联。MDCNN 基于关于 miRNA 和目标基因的异构信息网络 (HIN)，以元路径的形式对节点对之间的路径进行采样。然后将深度卷积神经网络（DCNN）学习到的节点特征和路径特征拼接在一起，作为miRNA-靶基因的表征，预测miRNA-靶基因的相互作用。实验结果表明，通过五重交叉验证，MDCNN 的性能在多个验证指标上优于其他方法。我们设置了一个消融研究来确定 miRNA 相似性和目标基因相似性的必要性，以提高 MDCNN 的预测能力。hsa-miR-26b-5p 和 CDKN1A 的案例研究进一步表明 MDCNN 可以成功预测潜在的 miRNA-靶基因相互作用。

图形摘要

更新日期：2021-06-25

点击分享查看原文

点击收藏

阅读更多本刊最新论文