Fig. 2.

Creating data sets of ICIs and NICIs and the corresponding data sets of noninteracting pairs. For the set of 8,695 interacting pairs, we generated a random set of 5,217,000 pairs. We chose this number to follow the rate of 1:600, based on the estimation of 30,000 interacting pairs and a total of 1.8 × 10⁷ possible protein pairs in S. cerevisiae. The interacting pairs included 1,466 pairs in complexes. The random pairs included 17,800 pairs where both proteins participate in the same complex. The latter were used as the data set of protein pairs in complexes that are not known to interact [random intracomplex pairs (RICPs)]. The rest of the interacting pairs included 7,229 pairs, and the rest of the random pairs included 5,199,200 pairs (a ratio of 1:719). In the analysis, because we used only 1,995 known interacting pairs that were the most reliable, we also reduced the random set accordingly to keep the same ratio of 1:719 between interacting and noninteracting pairs. As a result, the set of noninteracting protein pairs not in complexes included 1,434,405 pairs that were derived randomly from the rest of 5,199,200 random pairs [random non-intracomplex pairs (RNICPs)].