In this issue

When all else fails, put your heads together pp. 5243–5255 Membrane proteins use more than 25% of the coding capacity of most typical genomes. Type I membrane receptors are a diverse set of enzymes, receptors, channels, and structural elements that comprise about half of the total. The other half, type II receptors, are the members of the G protein coupled receptor (GPCR) family. Together they add up to approximately 1000 members in human cells. The number of messages conveyed by binding interactions is 10‐ to 100‐fold higher judging by the number of proteins that can bind to epidermal growth factor receptor (EGFR). Qi et al. worked out a procedure for predicting which membrane proteins could be expected to exhibit specific interactions. Into the mix went biological data including direct and indirect binding information and a goodly portion of human expertise. The output was computational predictions and global overviews that could be experimentally validated.Man to mouse: Projecting interactions pp. 5256–5266 On an evolutionary scale, the distance between men and mice is slight so we might expect that distances between regulatory and functional network components are also slight. Shin et al. looked at this question because, if true, it would make development of mouse‐based human model systems much quicker. There is a dearth of public mouse protein–protein interaction (PPI) information to compare to, and hence the researchers inferred two mouse PPIs in a two‐step procedure. ALL is derived from five human–mouse PPIs and CORE is refined from ALL. CORE is filtered for conserved protein domains as well as sequences. Upon comparing the inferred structures to the originals, there was little agreement – 1500 out of 42 000 were common to all. Although the networks projected were thin in terms of connections per node, they were good in terms of proximity of similarity of function.Caution! Deep water ahead. Frequent shark sightings pp. 5296–5302 No, the above was not a sign on an Australian beach I saw on my holiday. And, no, it is not a sign you would see as you approach “Auto Row” and all of its used car salesmen in a San Francisco suburb. I suggest editors stamp it on all “yeast‐two‐hybrid (Y2H)” manuscripts sent out for review. Rajagopala et al. decided to tackle the task of determining the sources of variance in the protein–protein interaction assay. They found that any one form of Y2H missed more than 75% of expected interactions. This was despite offering the test a variety of forms of bait, prey, vectors, copy numbers, reporters, vector origins and host species, for a start. Ultimately the interaction mapping should be predictable enough to serve as an internal quality check, a goal of these researchers.