Topology prediction of membrane proteins: how distantly related homologs come into play

The first atomic-resolution structure of a membrane protein was solved in 1985. After 25 years and 213 more unique structures in the database, we learned some remarkable biophysical features that thanks to computational methods help us to model the topology of membrane proteins (White 2009). However, not all the features can be predicted with statistically relevant scores when few examples are available (Oberai et al. Protein Sci 15: 1723–1734, 2006). Too often the notion that similar functions are supported by similar structures is expanded far behind the limits of a safe sequence identity value (>50%) to select templates for modeling the membrane protein at hand. To select proper templates we introduce a strategy based on the notion that remote homologs can have a role in determining the structure of any given membrane protein provided that the two proteins are co-existing in a cluster. Sequences are clustered in a set provided that any two sequences share a sequence identity value ≥ 40% with a coverage ≥ 90% after cross-genome comparison. This procedure not only allows safe selection of a putative template but also filters out spurious assignments of templates even when they are generally considered as the structure reference to a given functional family. The strategy also can play a role in indicating which membrane protein sets still would be worthwhile a structural investigation effort. Possibly when more membrane proteins will be available, the clustering system will allow fold coverage of the membrane protein universe.