基于靶标miRNA的药物重定位预测算法研究

Drug repositioning refers to finding new indications for existing drugs. Current research on drug repositioning mainly takes proteins as drug targets. However, the most recent studies from miRNA pharmacogenomics indicate that drugs can also target miRNAs and regulate their expressions to treat diseases, which provides a new perspective for drug repositioning. Information retrieval suggests that few studies have been conducted via miRNAs for drug repositioning in academic circle. Based on the newly discovered target miRNAs, this research aims to deepen the study of drug repositioning. The main content of this research includes: (1) developing non-negative matrix factorization-based algorithms to extract features from drugs, miRNAs and diseases to calculate their similarity; (2) based on the similarity measure, devising algorithms, by combining semi-supervised learning algorithms and recommendation algorithms, to predict miRNAs targeted by drugs and to predict the effects of inappropriate expressions of miRNAs on diseases, which could offer guidance for investigating the mechanism of action of drugs and the molecular mechanism of diseases; (3) constructing drug-miRNA-disease heterogeneous networks and proposing algorithms, based on the cause-and-effect principle, to predict drug-disease associations to finally achieve the goal of comprehensive drug repositioning.

药物重定位是指挖掘现有药物的新用途。当前药物重定位研究主要将蛋白质作为药物靶标进行分析。然而miRNA药物基因组学最新研究表明药物还可以靶向miRNA，调控其表达以达到治疗疾病的效果。这将为药物重定位提供新的依据。文献检索发现学术界对miRNA视角下的药物重定位研究较少。本课题拟基于miRNA这类全新的靶标深入研究药物重定位技术。主要研究内容有：(1)设计基于非负矩阵分解的特征提取算法抽取药物、miRNA和疾病的有效特征并依此计算这三类对象的相似性。(2)基于相似性度量，设计融合半监督学习算法和推荐算法等的二元关系预测算法以挖掘药物作用的靶标miRNA，预测miRNA异常表达导致的疾病。为探究药物的作用机理和疾病的致病机理提供依据。(3)构建药物-miRNA-疾病异构网络，研究基于因果关系原则下的预测算法挖掘药物-疾病关系，最终实现大规模药物重定位的目的。

越来越多的研究表明miRNA参与生命活动的多个重要进程，miRNA异常表达会引发多种复杂疾病。将药物靶向miRNA，调控其表达可以达到治疗疾病的效果。本项目基于miRNA这类全新靶标深入研究了药物重定位技术。课题组研究了疾病相关miRNA挖掘中的miRNA特征选择问题，研究结果表明miRNA序列，miRNA表达谱，miRNA靶向基因本体注释和miRNA关联疾病的MeSH均是疾病相关miRNA挖掘中的有效特征。针对疾病相关miRNA挖掘中负样本缺失问题，课题组设计了融合K-means和Rocchio分类的负样本筛选算法，结果表明该算法可选出更为可靠的负样本用于miRNA-疾病关联预测。课题组从典型关联分析、图注意力网络和图卷积网络角度分别设计了疾病相关miRNA挖掘算法，还基于靶标miRNA等信息设计了药物-疾病关联预测算法用于药物重定位筛选。课题组最后设计了多模态网络下融合图自动编码和注意力机制的算法用于药物敏感性预测。交叉验证实验结果表明上述关联预测算法可提升预测精度，案例分析实验结果验证了上述关联预测算法在真实场景下的有效性。本项目成果可为疾病病理研究和药物重定位筛选提供有益指导。在本项目资助下，课题组在Briefings in Bioinformatics、BMC Bioinformatics和BMC Medical Genomics等国际重要生物信息学期刊发表学术论文11篇，培养硕士研究生10名，其中毕业4名。课题组完成了项目预期目标要求。