ITF2357 (Givinostat) br The PRAD expression data sets were d
The PRAD ITF2357 (Givinostat) data sets were downloaded from the TCGA data portal. The RPM of miRNAs and the FPKM of mRNAs were normalized before our analysis. Similarly with the NEST, we estimated the essenti-ality of a node as the weight by summing of the expression level of its neighbors and itself. By speculating that the gene essentiality score would be altered under different conditions, we picked up those genes with significant essential score alterations in the tumor/normal by t-test.
Further, to validate whether the Jaccard similarity coefficient is reliable in our triple network, we compared the top-ranked 3596 mRNAs set by our method and NEST. KEGG pathways enrichment analysis was performed by GeneCodis3 .
Next, we investigated the pathways of the selected ESA miRNAs and DE miRNAs by using the DIANA miRPath Web server . Here, we used ‘TarBase’ to extract miRNA target genes and selected ‘pathways union’ to merge the results. For each miRNA, FDR <0.05 was used as a cutoff to identify the significant pathways. Meanwhile, we constructed a machine learning model, Gradient Boosting (GBT) Classifier , to estimate the predictive performance of selected ESA miRNAs or mRNAs on the tumor/normal samples. 10-fold cross-validation was performed.
Finally, feed-forward loops (FFLs) analysis be adopted to elucidate how these ESA genes in the FFLs participate in the cancer-related biological processes. Only FFLs comprised of ESA genes and with 1 DE genes would be considered in our analysis. Here, we mainly discussed the function of those genes in the FFLs. Then, we performed literature search on these mRNAs and miRNAs. The overview of the study was shown in Fig. 2.
3.1. Essentiality estimation based on miRNA-TF-mRNA co-regulatory network
In Our previous study, the NEST gene essentiality alterations were found performed well on identifying the abnormal genes across different cancer types . However, in NEST the PPI score  is directly used to weight the gene association, which is not suitable to the miRNA-TF-mRNA triple co-regulatory network. Here, we designed a method to identify essential gene by using Jaccard similarity coefficient  to weight the gene association. For node A and B, it is defined as:j=ðj þ j j.
To verify our methods, particularly the reliability of the Jaccard similarity coefficient, a comparison was carried out between our method and NEST. The mRNA and miRNA expression values were derived from The Cancer Genome Atlas (TCGA) data portal. The protein-protein in-teractions were downloaded from STRING database v9.1  by keeping those with confidence score > 0.6. For both methods, the gene essenti-ality scores were estimated by following the reported in Jiang et al. Only mRNAs were considered in the NEST, we, therefore, compared the top-ranked 3596 mRNAs by NEST and our method. There are 2212 (~62%) mRNAs detected by both methods. The mRNA sets enrichment analysis was additionally conducted by GeneCodis3 . Both mRNA sets are marked by the biological processes of cancer or cancer-related signaling pathways (Fig. 3a), such as Focal adhesion, Wnt signaling pathway, Regulation of action cytoskeleton, Pathways in cancer, ECM-receptor interaction and Prostate cancer. Their highly similar per-formance on identifying the essential genes suggested the Jaccard simi-larity method would be reliable in this task.
Fig. 3. The KEGG pathways enrichment analysis on ESA mRNAs and miRNAs in co-regulatory network and STRING network. (a) The significantly enriched pathways for ESA mRNAs in co-regulatory network (green) and in the protein-protein interaction network (orange). (b) The bar chart of enriched pathways of DE mRNAs. (c, d) The bar chart of enriched pathways of the top-ranked 100 ESA/DE miRNAs. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)