基于半监督弱标记的miRNA-放射病关联预测算法研究

As the important regulatory factor of life system, miRNAs regulate the gene expression by different ways and maintain the homeostasis of organism such as development, differentiation and apoptosis in biological precesses. Aiming at the specific characteristic of radiation sickness therapy, the project tries to investigate miRNA regulatory mechanisms and function to explore biotherapy for radiation disease. Firstly, considering the local label associations and the global label associations in biological network, the miRNA regulatory module recognition algorithm is designed based on the universality and specificity of miRNA regulatory function. Secondly, fully considering different omics data, the project constructs direct and indirect miRNA-disease association network, and then computes the class labels of each miRNA to obtain the weakly multi-labeled training data. Finally, ensemble algorithm for predicting miRNA-disease associations based on semi-supervised learning is proposed. This project will be significant to incorporating information methods such as machining learning into the biology science, which can be potentially applied to guide clinical treatment of radiation sickness.

miRNA作为生命系统重要的调控因子，通过不同的形式对基因表达进行调控，维持生物体的发育、分化和凋亡等生物过程的内稳态。本项目针对放射性疾病治疗的特殊性，以miRNA调控机制和功能为研究对象，探索生物基因疗法。首先，miRNA调控功能具有广泛性和特异性，综合考虑生物网络中miRNA局部标签关联和全局标签关联，研究适合于挖掘miRNA功能性调控模块的算法，并分析miRNA调控功能的动态演化特性。其次，基于不同组学数据，构建miRNA-疾病直接和间接关联网络，计算miRNA的类别标签，进而得到多标签弱标记的训练数据集，并针对训练样本的特点，提出半监督学习算法，预测miRNA与疾病间的关联。本项目对研究机器学习等信息方法与生命科学的结合具有重要的理论意义，并在指导临床治疗放射病方面具有潜在的应用价值。

miRNA作为生命系统重要的调控因子，通过不同的形式对基因表达进行调控，维持生物体的发育、分化和凋亡等生物过程的内稳态。本项目针对放射性疾病治疗的特殊性，以miRNA调控机制和功能为研究对象，探索生物基因疗法。首先，综合考虑各类组学数据，建立miRNA的类别标签，进而得到多标签弱标记的训练数据集，提出半监督多标签学习算法SMRM，挖掘miRNA功能性调控模块，构建miRNA-疾病关联网络，并揭示miRNA调控功能的动态演化特性，实验结果证明了算法的有效性。其次，miRNA调控功能具有广泛性和特异性，综合考虑生物网络中miRNA局部标签关联和全局标签关联，对缺失标签进行补全，形成用于训练学习模型的miRNA标签矩阵，提出半监督弱标记算法GLCLBL，预测miRNA与疾病间的关联。实验结果表明，本文提出算法显著提升了预测准确率。本项目对研究机器学习等信息方法与生命科学的结合具有重要的理论意义，并在指导临床治疗放射病方面具有潜在的应用价值。