Multivariate competing endogenous RNA network characterization for cancer microRNA biomarker discovery: a novel bioinformatics model with application to prostate cancer metastasis.

BACKGROUND

MicroRNAs (miRNAs) are post-transcriptional regulators with potential as biomarkers for cancer management. Data-driven competing endogenous RNA (ceRNA) network modeling is an effective way to decipher the complex interplay between miRNAs and spongers. However, there are currently no general rules for ceRNA network-based biomarker prioritization.

METHODS AND RESULTS

In this study, a novel bioinformatics model was developed by integrating gene expression with multivariate miRNA-target data for ceRNA network-based biomarker discovery. Compared with traditional methods, the structural vulnerability in the human long non-coding RNA (lncRNA)-miRNA-messenger RNAs (mRNA) network was comprehensively analyzed, and the single-line regulatory or competing mode among miRNAs, lncRNAs, and mRNAs was characterized and quantified as statistical evidence for miRNA biomarker identification. The application of this model to prostate cancer (PCa) metastasis identified a total of 12 miRNAs as putative biomarkers from the metastatic PCa-specific lncRNA-miRNA-mRNA network and nine of them have been previously reported as biomarkers for PCa metastasis. The receiver operating characteristic curve and cell line qRT-PCR experiments demonstrated the power of , and as novel candidates for predicting PCa metastasis. Moreover, PCa-associated pathways such as prostate cancer signaling, signaling, and signaling were significantly enriched by targets of identified miRNAs, indicating the underlying mechanisms of miRNAs in PCa carcinogenesis.

CONCLUSIONS

A novel ceRNA-based bioinformatics model was proposed and applied to screen candidate miRNA biomarkers for PCa metastasis. Functional validations using human samples and clinical data will be performed for future translational studies on the identified miRNAs.