Figure 1. Network inference method flowchart.

Phase 1 consists of constructing a scaffold network using a set of chosen genes thought to sufficiently represent the cardiogenesis process. Two independent metrics based on the interaction frequencies generated by the inference algorithm, the confidence metric, and gene ontology, the semantic similarity metric, are used to filter the network and remove spurious interactions. Phase 2 involves expansion of this scaffold network using a cluster expansion technique to produce a more complete network that best characterizes the regulatory interactions during cardiogenesis as inferred from the data. The gene interactions are further prioritized using a gene-profile uniqueness metric, the cluster product, to generate an experimentally realizable set of predictions.