College of Information Science and Engineering, Zaozhuang University, Zaozhuang 277122, China.
School of Computer Science and Technology, China University of Mining and Technology, Xuzhou 221008, China.
Brief Funct Genomics. 2023 Nov 10;22(5):453-462. doi: 10.1093/bfgp/elad010.
A large number of experiments show that the abnormal expression of miRNA is closely related to the occurrence, diagnosis and treatment of diseases. Identifying associations between miRNAs and diseases is important for clinical applications of complex human diseases. However, traditional biological experimental methods and calculation-based methods have many limitations, which lead to the development of more efficient and accurate deep learning methods for predicting miRNA-disease associations.
In this paper, we propose a novel model on the basis of adaptive deep propagation graph neural network to predict miRNA-disease associations (ADPMDA). We first construct the miRNA-disease heterogeneous graph based on known miRNA-disease pairs, miRNA integrated similarity information, miRNA sequence information and disease similarity information. Then, we project the features of miRNAs and diseases into a low-dimensional space. After that, attention mechanism is utilized to aggregate the local features of central nodes. In particular, an adaptive deep propagation graph neural network is employed to learn the embedding of nodes, which can adaptively adjust the local and global information of nodes. Finally, the multi-layer perceptron is leveraged to score miRNA-disease pairs.
Experiments on human microRNA disease database v3.0 dataset show that ADPMDA achieves the mean AUC value of 94.75% under 5-fold cross-validation. We further conduct case studies on the esophageal neoplasm, lung neoplasms and lymphoma to confirm the effectiveness of our proposed model, and 49, 49, 47 of the top 50 predicted miRNAs associated with these diseases are confirmed, respectively. These results demonstrate the effectiveness and superiority of our model in predicting miRNA-disease associations.
大量实验表明，miRNA 的异常表达与疾病的发生、诊断和治疗密切相关。鉴定 miRNA 与疾病之间的关联对于复杂人类疾病的临床应用非常重要。然而，传统的生物实验方法和基于计算的方法存在许多局限性，这导致了更高效、更准确的深度学习方法的发展，以预测 miRNA-疾病关联。
在本文中，我们基于自适应深度传播图神经网络提出了一种新的模型来预测 miRNA-疾病关联（ADPMDA）。我们首先基于已知的 miRNA-疾病对、miRNA 综合相似性信息、miRNA 序列信息和疾病相似性信息构建 miRNA-疾病异质图。然后，我们将 miRNA 和疾病的特征投影到低维空间中。之后，利用注意力机制聚合中心节点的局部特征。特别是，采用自适应深度传播图神经网络来学习节点的嵌入，从而能够自适应地调整节点的局部和全局信息。最后，利用多层感知机对 miRNA-疾病对进行评分。
在人类 microRNA 疾病数据库 v3.0 数据集上的实验表明，ADPMDA 在 5 折交叉验证下的平均 AUC 值达到 94.75%。我们进一步对食管癌、肺癌和淋巴瘤进行了案例研究，以确认我们提出的模型的有效性，分别有 49、49 和 47 个与这些疾病相关的 top50 预测 miRNA 得到了验证。这些结果表明了我们的模型在预测 miRNA-疾病关联方面的有效性和优越性。
