MicroRNA as an integral part of cell communication: regularized target prediction and network prediction.

MicroRNAs, gene encoded small RNA molecules, play an integral part in gene regulation by binding to target mRNAs and preventing their translation. The prediction of microRNA- mRNA binding sites and the resulting interaction network are essential to understand, and thus influence, regulation of a genetic information flow inside the living organism. Numerous algorithms have been proposed based on various heuristics; however the predictions often vary considerably. In this proposal we will extend a physical model for the binding of microRNAs to the corresponding target and establish an extended set of features influencing binding probabilities. We will be faced with the challenge of (i) too many features and (ii) few known interactions on which to train any prediction algorithm. This problem will be solved by using (i) information-theoretical criteria for feature reduction, (ii) regularization, (iii) application of the Infomax approach to guarantee minimal loss of information after dimension reduction, and (iv) experimental validation of theoretical predictions using a novel test system. This strategy will allow (i) statistical analysis of the predicted microRNA-mRNA hypergraph, (ii) characterization of network motives and hierachies, (iii) identification of missing links and (iv) removal of false interactions.