Validation procedure for the STAR model: In order to demonstrate the feasibility of the STAR model, we designed a recapturing scheme. We removed one link at a time from the network and examined whether this link can be recaptured by application of the STAR model to this reduced network. This figure explains the validation procedure using the MCM1 – CLB2 link as an example. (a) We first applied the USA to the local (STAR-like) network of MCM1 to find experimental conditions in which the core input genes are over-expressed or under-expressed. We then evaluated the correlations between all the targets of MCM1 under these (UTM) conditions, and removed targets whose correlation with other members of the core set are insignificant. We recursively applied the USA in order to remove MCM1 target genes that are weakly correlated with any other of the MCM1 target genes. At each step of the recursive elimination, we identified a centroid gene that has the largest number of highly correlated genes. We eliminated genes that are not highly correlated with the centroid gene. We continued these iterations until all the remaining target genes are highly correlated with each other. CLB2 was identified as the centroid in the last iteration. (b) In the next step we removed the link MCM1→CLB2 associated with the centroid gene, and applied the STAR model described in Figure l(b). As shown, CLB2 reappeared in the UTM generated from the application of the STAR model to a core set of input genes that excludes CLB2. Moreover, the matching score between the PWM of MCM1 and the promoter sequence of CLB2 was high. Overall, we recaptured 36% of the centroid genes by applying this procedure to all the multi-target local networks in the literature-driven gene regulatory network.