Labeling negative examples in supervised learning of new gene regulatory connections

摘要: Supervised learning methods have been recently exploited to learn gene regulatory networks from expression data. The basic approach consists into building a binary classifier feature vectors composed by levels of set known connections, available in public databases or literature. Such is then used predict new unknown connections. The quality the training plays crucial role such an inference scheme. In classification should be positive and negative examples, but Biology literature only collected information whether two genes interact. Instead, counterpart usually not reported, as Biologists are aware state interacting. The over presence topology motifs currently networks, as, feed-forward loops, bi-fan clusters, single input modules, could drive selection reliable examples. We introduce, discuss, evaluate number heuristics that exploits network Escherichia coli Saccharomyces cerevisiae.