Figure 1.
Overview of the proposed approach.
(A) Collect gene expression and miRNA expression data sets from paired tumor samples, and calculate log2 ratios between tumor samples and normal samples. (B) Construct gene-sample modules (GSM) from a differentially expressed gene expression matrix using a biclustering algorithm, which allows duplications of genes and samples in multiple modules. (C) Add genes to GSM using gene-gene interactions, if the included genes increase the average PCC values among genes in the module. (D) Construct gene-miRNA modules (GMM) by selecting gene-regulating miRNAs in GSM. Use a Gaussian Bayesian network and the BIC score to evaluate the relationship between genes and miRNAs. (E) To determine the functional relevance of the modules, test whether the genes from the modules are enriched for specific biological functions or signaling pathways. To validate that modules are related to a specific cancer, check that the genes and miRNAs are related to the specific cancer.
Figure 2.
Performance comparison of gene-miRNA modules for ovarian cancer.
For ovarian cancer, we compared the performance of gene-miRNA modules generated from four cases: SCC with GGI information, SCC without GGI information, PCC with GGI information, and PCC without GGI information. For all cases, the x-axis presents different percentages of candidate miRNAs (T%) among all miRNAs when constructing gene-miRNA modules. For each case, the number of modules (A), the ratios of cancer genes (B), the ratios of ovarian cancer genes (C), the ratios of ovarian cancer miRNAs (D), the average number of enriched pathways (E), and the ratios of modules enriched with at least one pathway (F) are shown.
Table 1.
Cancer genes, ovarian cancer genes and ovarian cancer miRNAs for selected modules.
Table 2.
miRNAs regulate genes in ovarian cancer modules.
Table 3.
Experimentally validated gene-miRNA interactions with strong evidence from miRTarbase in ovarian cancer modules.
Figure 3.
Regulations among genes, miRNAs, and TFs in ovarian cancer modules.
For three ovarian cancer modules-22 (A), 8 (B), and 33 (C)-the expression values of genes, miRNAs, and TFs are shown. Arrows represent genes and miRNAs regulated by TFs or other miRNAs. Genes and miRNAs are members of each module, but TFs do not belong to the modules.
Figure 4.
Regulations among genes, miRNAs, and TFs in GBM modules.
For two GBM modules, 11 (A) and 5 (B), the expression values of genes, miRNAs, and TFs are shown. Arrows represent genes and miRNAs regulated by TFs or other miRNAs. Genes and miRNAs are members of each module, but TFs do not belong to the modules.
Table 4.
Ovarian cancer modules with enriched pathways.
Figure 5.
Network presentation of module 22 in ovarian cancer.
In this network, diamonds represent miRNAs: sky-blue nodes for ovarian cancer miRNAs from the HMDD database, pink nodes for ovarian cancer miRNAs supported by the literature, and yellow nodes for the remaining miRNAs. Genes are represented by circles: pink nodes for ovarian cancer genes validated by the literature, green nodes for ovarian cancer genes validated by the DDOC database, orange nodes for cancer genes, and white nodes for the remaining genes. A blue solid line indicates that the MCC value between a gene and a miRNA is larger than 0.3. A purple line indicates that the linked genes are enriched together with at least one function. For example, COL6A1, COL5A3, THBS2, FN1, COL1A1, COL5A1, COLA1A, and COL3A1 are enriched with at least one function together (ECM receptor pathway or Focal adhesion pathway). Table S17 presents PubMed identifiers for ovarian cancer genes in pink nodes.
Figure 6.
Network presentation of module 8 in ovarian cancer.
The description of this network is the same as in Fig. 5 except that red lines are used to represent two enriched pathways (complement and coagulation cascades pathway, and TGF signaling pathway).
Figure 7.
The description of this network is the same as in Fig. 5, except that green nodes indicate GBM genes validated by two articles [34, 35], and pink nodes indicate GBM genes validated by the literature in PubMed. Table S17 presents PubMed identifiers for GBM genes.
Figure 8.
Expression levels of ovarian cancer subtype marker genes.
(A) Heat map of the means of marker gene expression levels for 32 ovarian cancer modules. Red indicates overexpression of genes, and green indicates underexpression of genes. (B) Expression levels of marker genes of selected modules. Blue bars represent marker genes that determine the subtype and red bars represent other subtype marker genes.
Figure 9.
Expression levels of GBM subtype marker genes.
(A) Heat map of the means of marker gene expression levels for 54 GBM modules. Red indicates overexpression of genes, and green indicates underexpression of genes. (B) Expression levels of marker genes of selected modules. Blue bars represent marker genes that determine the subtype, and red bars represent other subtype marker genes.
Figure 10.
Comparison of modules identified using our approach and the NMF approach using ovarian cancer data. (A) The ratio of modules with at least one enriched function or pathway. (B) The average number of enriched functions in the identified modules. (C) The average ratios of cancer genes, ovarian cancer genes, and ovarian cancer miRNAs in the modules.