A Bayesian Framework to Improve MicroRNA Target Prediction by Incorporating External Information.

MicroRNAs (miRNAs) are small regulatory RNAs that play key gene-regulatory roles in diverse biological processes, particularly in cancer development. Therefore, inferring miRNA targets is an essential step to fully understanding the functional properties of miRNA actions in regulating tumorigenesis. Bayesian linear regression modeling has been proposed for identifying the interactions between miRNAs and mRNAs on the basis of the integrated sequence information and matched miRNA and mRNA expression data; however, this approach does not use the full spectrum of available features of putative miRNA targets. In this study, we integrated four important sequence and structural features of miRNA targeting with paired miRNA and mRNA expression data to improve miRNA-target prediction in a Bayesian framework. We have applied this approach to a gene-expression study of liver cancer patients and examined the posterior probability of each miRNA-mRNA interaction being functional in the development of liver cancer. Our method achieved better performance, in terms of the number of true targets identified, than did other methods.