TCNLDA: Prediction of lncRNA-Disease Associations Based on Autoencoder and Temporal Convolutional Network

研究表明长非编码RNA (long non-coding RNA, lncRNA)影响着许多疾病的生物学过程，例如疾病的发生、传播、治愈等。因此，预测潜在lncRNA-疾病关联(lncRNA-disease associations, LDAs)对疾病的诊疗和治疗有着重要意义。本文提出一种新的深度学习方法预测LDAs，称为TCNLDA。首先，分别构建了lncRNA的功能相似性矩阵、高斯相似性矩阵和序列相似性矩阵，以及疾病的语义相似性矩阵和高斯相似性矩阵，并将其进行矩阵融合处理。然后，构建lncRNA-疾病对，并对其使用自编码器(Autoencoder, AE)进行特征提取。最后，将提取好的特征输入到时序卷积网络(Temporal Convolutional Network, TCN)中进行训练输出预测得分。两个数据集中，TCNLDA与其他模型进行了比较，结果显示TCNLDA优于其他LDAs预测方法。消融实验验证了TCNLDA中各部分的不可缺少性。案例研究进一步表明，TCNLDA在预测新型LDAs方面有着很好的实用性。Studies have shown that long non-coding RNA (lncRNA) influences the biological processes of many diseases, such as disease onset, spread, and cure. Therefore, predicting potential lncRNA-disease associations (LDAs) is important for disease diagnosis and treatment. In this paper, we propose a new deep learning method to predict LDAs, called TCNLDA. Firstly, the functional similarity matrix, Gaussian similarity matrix and sequence similarity matrix of lncRNAs, and the semantic similarity matrix and Gaussian similarity matrix of diseases are constructed and processed for matrix fusion, respectively. Then, the lncRNA-disease pairs are constructed and feature extraction is performed on them using autoencoder (AE). Finally, the extracted features were input into Temporal Convolutional Network (TCN) for training to output the prediction scores. In both datasets, TCNLDA was compared with other models, and the results showed that TCNLDA outperformed other LDAs prediction methods. Ablation experiments verified the indispensability of the components in TCNLDA. The case study further shows that TCNLDA has good utility in predicting novel LDAs.