Proteome network emulating models

The proteome network (or protein-protein interaction (PPI) network) of an organism represents each protein as a vertex and each pairwise interaction as an edge. In the past 10 years, we witnessed a significant amount of effort going into the development of the PPI networks and the computational tools for analyzing them. In particular, there have been several attempts to capture the topological features of PPI networks through random graph models, which have been successfully applied to the emulation of small-world networks, which are sparse, but highly connected. The available PPI networks have also been thought to have a small diameter with power-law degree distribution thus scale-free network emulators such as the Preferential Attachment Model have been investigated for the purposes of emulating PPI networks. The lack of success in this direction led to the development of further models, which either reject the scale-freeness of the PPI networks, such as the Geometric Random Network Model or guarantee scale freeness through means of expansion other than Preferential Attachment such as vertex (i.e., protein/gene duplication) - as in the case of the Pastor-Satorras Model or the more recent Generalized Duplication Model. In this study, we compare available PPI networks of various sizes with those generated by the random graph models and observe that the Generalized Duplication Model, with the right choice of the initial seed network, provides the best alternative in capturing all network feature distributions. One network feature distribution that remains difficult to capture, however, is the dense graphlet distribution: all available PPI networks seem to include (many) more dense graphlets such as cliques in comparison to the networks generated by all available models.