Interaction data was gathered from a number of different published high-throughput datasets and published databases [2–5 (link),27 (link),34 (link),45 (link),46 (link)]. Independent genomic features and Bayesian integration were used to eliminate noise from the dataset [23 (link),43 (link)]. Different datasets (e.g., the FYI [Vidal et al.] [21 (link)] or the DIP core [Eisenberg et al.] [44 (link)]) exhibit the same behavior (see Figures S1 and S4A). To avoid biases from large complexes (i.e., the ribosome and the proteasome), we repeated our calculations after removing both these complexes (see Figures S2 and S4B). The regulatory network was created by combining five different datasets [1 (link),2 (link),22 ,34 (link),35 (link),47 (link)]. We excluded DNA-binding enzymes (e.g., PolIII) from the regulatory network. The essential genes in yeast genome were determined experimentally through a PCR-based gene-deletion method [36 (link)]. The metabolic network was taken from the Kyoto Encyclopedia of Genes and Genomes (KEGG) [48 (link)] and all proteins that share a metabolite were considered linked. The genetic network data was downloaded from the GRID [49 (link)] and consists of several large-scale screens of genetic interactions [30 (link),50 (link)]. Expression data was taken from the Rosetta compendium expression dataset [51 (link)]. All datasets and the calculated betweenness of each protein node within these networks are available at http://www.gersteinlab.org/proj/bottleneck. Because most of these networks are far from complete, we will update the networks and, more important, the associated betweenness of each node as they grow in size in the future.
Free full text: Click here