A Computational Approach to Identifying Gene-microRNA Modules in Cancer

MicroRNAs (miRNAs) play key roles in the initiation and progression of various cancers by regulating genes. Regulatory interactions between genes and miRNAs are complex, as multiple miRNAs can regulate multiple genes. In addtion, these interactions vary from patient to patient and even among patients with the same cancer type, as cancer development is a heterogeneous process. These relationships are more complicated because transcription factors and other regulatory molecules can also regulate miRNAs and genes. Hence, it is important to identify the complex relationships between genes and miRNAs in cancer.
In this study, we propose a computational approach to constructing modules that represent these relationships by integrating the expression data of genes and miRNAs with gene-gene interaction data. First, we used a biclustering algorithm to construct modules consisting of a subset of genes and a subset of samples to incorporate the heterogeneity of cancer cells. Second, we combined gene-gene interactions to include genes that play important roles in cancer-related pathways. Then, we selected miRNAs that are closely associated with genes in the modules based on a Gaussian Bayesian network and Bayesian Information Criteria. When we applied our approach to ovarian cancer and glioblastoma (GBM) data sets, 33 and 54 modules were constructed, respectively. In these modules, 91% and 94% of ovarian cancer and GBM modules, respectively, were explained either by direct regulation between genes and miRNAs or by indirect relationships via transcription factors. In addition, 48.4% and 74.0% of modules from ovarian cancer and GBM, respectively, were enriched with cancer-related pathways, and 51.7% and 71.7% of miRNAs in modules were ovarian cancer-related miRNAs and GBM-related miRNAs, respectively. Finally, we extensively analyzed significant modules and showed that most genes in these modules were related to ovarian cancer and GBM.

微小RNA（miRNAs）通过调控基因，在多种癌症的发生与发展过程中发挥关键作用。基因与miRNAs之间的调控关系极为复杂，多条miRNAs可同时调控多个基因。此外，由于癌症发展具有异质性，这类调控关系在不同患者之间甚至在同一癌症类型的患者群体中均存在差异。而转录因子及其他调控分子亦可对miRNAs与基因进行调控，进一步加剧了这些关系的复杂性。因此，解析癌症中基因与miRNAs之间的复杂调控关系具有重要意义。

本研究提出一种计算方法，通过整合基因与miRNAs的表达数据以及基因间互作数据，构建能够表征上述调控关系的模块。首先，采用双聚类算法（biclustering algorithm）构建由部分基因与部分样本组成的模块，以纳入癌细胞的异质性特征；其次，结合基因间互作数据，将参与癌症相关通路的关键基因纳入模块；随后，基于高斯贝叶斯网络（Gaussian Bayesian network）与贝叶斯信息准则（Bayesian Information Criteria），筛选出与模块内基因密切相关的miRNAs。将该方法应用于卵巢癌与胶质母细胞瘤（glioblastoma, GBM）数据集后，我们分别构建出33个和54个模块。在这些模块中，分别有91%的卵巢癌模块与94%的GBM模块可通过基因与miRNAs间的直接调控，或是经由转录因子介导的间接关系得到合理解释。此外，分别有48.4%的卵巢癌模块与74.0%的GBM模块富集了癌症相关通路；模块内的miRNAs中，分别有51.7%为卵巢癌相关miRNAs，71.7%为GBM相关miRNAs。最后，我们对显著模块进行了全面分析，结果显示这些模块内的绝大多数基因均与卵巢癌和GBM相关。