基于潜在因子模型的环状RNA与疾病相关性研究

近年来，大量的研究表明共价闭环的环状RNA在生物过程中发挥着重要的功能。环状RNA的失调和突变与疾病有关。由于稳定的结构和抗降解性，环状RNA具有作为诊断性标志物的潜力。环状RNA与疾病相关性研究将有助于疾病诊断和治疗。我们从分子水平和系统水平上对生物数据使用表征学习来发掘生物特征，从而提高环状RNA与疾病相关性预测的准确度。在分子水平上，利用词向量模型挖掘环状RNA序列的生物特征，提取疾病的表型特征，引入生物信息减少对已知关联的过度依赖。在系统水平上，利用异构网络表征模型从多个不同类型的分子网络中学习环状RNA与疾病生物特征，多个角度补充环状RNA与疾病的生物信息。此外，针对组织特异性的多源异构网络，引入注意力机制从节点层次和元路径层次学习环状RNA与疾病生物特征，进一步细分环状RNA与疾病的相关程度。本项目将开发统一的数据查询、数据映射和环状RNA与疾病相关性研究软件平台。

Recently, a large number of studies indicate that circRNAs with covalently closed loops play important roles in biological processes. Dysregulation and mutation of circRNAs are related to diseases. Due to its stable structure and resistance to degradation, circRNA has the potential as a diagnostic biomarker. Therefore, research on the relation of circRNA-disease is helpful in disease diagnosis and disease treatment. We use representation learning on biological data to discover biological features at the molecular level and the system level for improving accuracy. At the molecular level, functional patterns of the circRNAs' sequences are minded by the word vector model. Meanwhile, the phenotypic patterns of diseases are extracted from disease ontology. These approaches not only introduce biological characteristics but also reduce over-relevance on known associations. At the system level, representation learning is used on the heterogeneous multi-source network to learn the biological features of circRNAs and diseases during the pathogenesis. As for the tissue-specific multi-source heterogeneous network, an attention mechanism is introduced to learn heterogeneous features. This project will develop a software platform for predicting circRNA-disease associations, data query and data mapping of circRNAs and diseases.