Deciphering microRNA-mediated regulation from perturbations in gene expression

by

Christina Leslie
Computational Biology Program
Memorial Sloan-Kettering Cancer Center

Friday, March 27, 2009
12:00 lunch
12:15 seminar
402 Walter Library

microRNAs (miRNAs) are an important class of small regulatory RNAs that are involved in post-transcriptional gene silencing. Intensive research within the past decade has revealed that miRNAs play critical roles in development, cell differentiation, and regulation of cell cycle and programmed cell death, as well as in the pathogenesis of cancer. A number of groups have used genomics approaches to study the function of miRNAs, including transfection of miRNAs into cell lines followed by microarray profiling of gene expression changes. Our interest is to use genome-wide data from miRNA transfection experiments to computationally decipher gene regulation by miRNAs.

We address two different questions in our study. First, we consider the crucial and still unsolved problem of predicting the targets of miRNAs. Previous studies have established that miRNAs recognize partially complementary sites in the 3' untranslated region of transcripts, with near-perfect complementarity required at the “seed” region of the miRNA. However, seed matches alone give poor specificity, and typical target prediction algorithms generate large numbers of false positives. We use genome-wide expression changes following miRNA transfections as training data for a new supervised learning algorithm, called mirSVR, to predict the efficiency of potential miRNA target sites. mirSVR uses a multivariate feature representation — including features of the predicted miRNA::site duplex, contextual sequence features, and predicted secondary structure accessibility — for potential sites in a support vector regression (SVR) approach. In a large panel of independent miRNA transfection experiments, mirSVR significantly outperforms leading miRNA target site prediction algorithms for the task of ranking genes that are down-regulated at the mRNA or protein level.

Second, we observed that in miRNA and siRNA transfections, a substantial fraction of genes are in fact up-regulated at the transcript or protein level, even though these small RNAs post-transcriptionally down-regulate their targets. While these changes have previously been dismissed as secondary effects, we computationally investigate a different hypothesis: namely, that miRNA and siRNA transfections globally disrupt gene regulation by the cell’s own miRNAs, possibly by saturation of protein machinery (e.g. RISC) needed to carry out gene silencing. We give strong statistical evidence in support of this hypothesis through analysis of over 150 mi/siRNA experiments, with implications for siRNA screens, miRNA target prediction, and small RNA therapeutics.

---