Predicting MicroRNA targets by integrating sequence and expression data in cancer

Gene regulation is a key factor in gaining a full understanding of molecular biology. microRNA (miRNA), a novel class of non-coding RNA, has recently been found to be one crucial class of post-transactional regulators, and play important parts in cancer. One essential step to understand the regulatory effect of miRNAs is the reliable prediction of their target mRNAs. Typically, the predictions are solely based on sequence information, which unavoidably have high false detection rates. Here we develop a new algorithm called HCTarget, which predict miRNA targets by integrating the typical algorithm and the paired expression profiles of miRNA and mRNA. HCTarget formulates a linear model to characterize the relationship between mRNA and miRNA, and use a Markov Chain Monto Carlo algorithm to learn the target probabilities. When applying HCtarget to the expression data in multiple myeloma, we predict target genes for ten cancer related miRNAs. The experimental verification and a loss of function study of hsa-miR-16 validate our predictions. Compared with the previous approaches, our target sets have increased functional enrichment. Meanwhile, our predicted target pair hsa-miR-19b and SULF1 plays an important role in multiple myeloma. Therefore, HCtarget is a reliable and effective approach to predict miRNA target genes, and could improve our comprehensive understanding of gene regulation.