Figure 6.

Specificity of interactions. (a) A hypothetical subnetwork for non-specific interaction between proteins containing two domains: each protein containing domain A interacts with each protein containing domain B. Detecting such interactions is easy for all four methods: Association, EM, DPEA, and PE. (b) A hypothetical subnetwork for highly specific interactions between proteins containing domain A and proteins containing domain B. Since only a small number of interactions actually occur, out of all possible interactions between pairs of proteins containing domain pair {A, B}, detecting such specific interactions is difficult for the EM and the Association methods, but not for the DPEA and the PE methods. (c) Hypothetical subnetwork for highly specific interactions in the context of multidomain proteins. PE will attribute these interactions to domain pair {A, B}, as it requires prediction of one interaction {A, B} to justify three protein-protein interactions. On the other hand, the association and the EM method will assign higher probability to domain pairs {X, X'}, {Y, Y'}, and {Z, Z'}, as it is beneficial to assign higher probabilities to interactions involving rare domains, that is, X, Y, and Z.