Integrative computational analysis of micro RNA and mRNA expression profiles in human cancer.

摘要

Mature microRNAs (miRNAs) are short (19-24 nt), non-protein-coding ribonucleic acids that play very important roles in the regulation of gene expression in animals and plants. miRNAs mainly bind to the 3' untranslated regions of target mRNAs to cause translational blockade or transcript degradation. Although miRNAs have been implicated in growing number of diseases, their protein targets and the specific biological functions of these targets remain largely unknown. Computational prediction of miRNA targets provides an alternative approach to assign biological functions. Although the experimental validation of miRNA target genes increases dramatically, majority of miRNA targets are still unknown and bioinformatic algorithms remain the key means of predicting putative miRNA targets. The principles of miRNA target predictions are based on sequence complementary, conservation across species, thermodynamic stability, site accessibility and inverse relationship between the expression profiles of miRNAs and predicted target mRNAs.;Here we use partial least square (PLS) and sparse partial least square (SPLS) methods to predict miRNA targets from miRNA and mRNA microarray data. Based on the inverse relationship between miRNA and mRNA, we selected two sets of differentially expressed miRNAs and mRNAs from human colon cancer microarray data. The first set consisted of 71 upregulated mRNAs and 31 downregulated miRNAs and the second set consisted of 56 downregulated mRNAs and 2 upregulated miRNAs. Using PLS and SPLS methods, we detected significant inverse interactions/associations between miRNA and mRNA. Then we compared these miRNA target genes with the four other widely used miRNA target prediction programs: TargetScan 5.1, PicTar, miRanda and miRBase. We identified a set of miRNA targets predicted by PLS and/or Sparse PLS that were also predicted by TargetScan5.1, PicTar, miRanda and miRBase through union of them or intersection combinations. We also used our predicted miRNA target genes to explore miRNA-mediated biological networks or pathways in human cancer.