人类microRNA识别及其与疾病关联预测方法研究

microRNA (miRNA) identification and function prediction is one of popular problems of bioinformatics. The project will develop miRNA computational identification method aiming at the human genome-wide data. Furthermore, the method of predicting association between miRNAs and human diseases will be studied to explore miRNA function in diseases. Firstly, we investigate the feature selection of pre-miRNA, pre-miRNA classification based on class imbalance, screening pre-miRNA candidates, and miRNA position prediction, which is helpful for improving the efficiency of identificating miRNAs over the human genome-wide data. Secondly, we present the miRNA functional similarity measurement which fully considering the semantic description of disease and disease phenotype information. Moreover, the threshold selection based on clustering coefficient is used to construct a reliable miRNA functional similarity network. Finally, the new prediction method based on random walks is proposed by incorporating the functional similarity information between nodes, the structural information of network, and the prior information of nodes. This method can predict the miRNAs associated with given human disease such as heart failure. The project has great significance for exploiting the new computational identification and prediction methods of miRNA and incorporating algorithm theories into the life science. It also has potentially application value in exploring mechanisms of human disease.

microRNA（miRNA）的识别及其功能预测是当前生物信息学领域的热点问题。本项目发展面向人类全基因组的miRNA计算识别方法，并研究miRNA与疾病关联的预测方法，以探索其在疾病中行使的功能。首先，以机器学习方法为基础，研究miRNA前体的特征选择、面向类别不平衡的前体分类、前体候选的筛选和miRNA成熟体位置预测，以提高针对人类全基因组的miRNA识别效果。然后，提出一种充分考虑疾病语义描述和疾病间表型信息的miRNA功能相似性计算方法；进而采用基于聚类系数的阈值选择方法，构建可靠的miRNA功能网络。最后，结合结点间相似性信息、网络结构信息和先验信息，提出新的基于随机游走的预测算法，有效预测与心力衰竭等给定人类疾病关联的miRNA。本项目对开拓新的miRNA的生物信息学识别与预测方法、研究算法理论与生命科学的结合都有重要的理论意义，并在探索人类疾病发生机理方面具有潜在的应用价值。

本项目以机器学习等信息学方法为基础，充分考虑miRNA的分子生物学特性，深入开展了miRNA识别及其与人类疾病关联预测的研究。在面向人类全基因组的miRNA的识别方面，首先针对miRNA前体特征冗余问题，提出有代表性特征子集的选择方法，为建立具有高预测性能的miRNA前体分类模型奠定基础。其次，针对真、假miRNA前体存在的类别不平衡问题，提出了基于集成学习的miRNA前体分类方法，从而得到具有高敏感性和特异性的前体分类预测模型。第三，结合miRNA前体的能量和热力学稳定的特点，给出了从全基因组范围内众多miRNA前体候选中筛选有潜力miRNA前体的方法。第四，针对预测得到的miRNA前体候选，结合miRNA成熟体生源学特点，提出了在前体候选中准确预测成熟体位置的方法。本项目有效提高了人类全基因组的非同源和物种特异miRNA前体和成熟体的识别效果，有助于改善生物实验发现真miRNA前体和成熟体的效率。此外，在预测miRNA与疾病的关联方面，一方面进行多信息融合并充分挖掘疾病描述的语义，提出了准确的miRNA功能相似性评估方法；另一方面，结合功能相似性信息、网络结构信息、结点先验信息，提出了新的基于随机游走的疾病miRNA预测方法。本项目系统预测了与心脏衰竭等15种人类常见疾病相关的候选，推动了后续miRNA在疾病发生发展过程中调控作用机制方面的研究。