Construction of signaling networks with incomplete RNAi data

Methods for constructing signaling networks from reference networks and single gene knockdown RNAi experiments have been proposed in recent years. All of these studies assume that the RNAi data is complete. However, RNAi experiments are usually noisy and more importantly have a considerable amount of missing data (i.e., a subset of the gene knockdowns is missing). In this paper, we address the signaling network construction problem with incomplete RNAi data. We develop two new methods for constructing a network topology which is closest to the reference network and consistent with the given incomplete RNAi data. Our experiments on real and synthetic datasets demonstrate that these methods produce accurate results and they are efficient. For real Wnt networks, our methods produce results with high accuracy in less than 100 ms.