Subnetwork evaluation is performed by comparing phenotype associated MCNs versus random MCNs with the same number of proteins/genes. In particular, we are interested in whether the average number of nodes per component is greater than random. This parameter was shown to be the best one to compare and detect MCNs of interest as it can distinguish random MCNs from known pathways and molecular complexes associated MCNs.

To perform this statistical analysis it has been calculated 10000 random lists for the size range between 10 and 200. For each of these random lists, Network Miner obtains their MCNs and calculates the average number of nodes per component (connected set units). Based on the features of each random network generated (number of components and number of proteins forming it), we can obtain an estimate of a p value simply by ranking the feature of the network of interest on them. Thus, p values of comparisons with alternative greater can be obtained.

Since Network Miner offers the possibility of calculating the MCNs with and without incorporating an external node (proteins not present in the original list but can link two or more proteins of the list), random mcn had been calculated using the two possibilities.