Figure 5

Arbitrary docking also targets alternative interfaces. A: Example of a protein for which the prediction of the biological interface from arbitrary docking appears to fail. The Colicin-E7 immunity protein forms a dimer with the E7 protein in the bound structure of the docking data set (7CEI_r [49], unbound form 1UNK [51]). This protein also forms an octamer involving four Colicin-E7 immunity proteins and four E7 proteins in structure 2JAZ [50]. The surface of Colicin-E7 immunity protein is colored according to the number of docking hits obtained with the 25 shortest probes and 10 docking models, highlighting the fact that arbitrary docking targets a region involved in the interface of the octamer. B: Description of the procedure used to detect multiple interfaces with the PiQSi database. Starting from the 198 proteins in the target set, we obtain 80 proteins for which PiQSi retrieves homologues with more than 35% identity involved in protein complexes. These 80 proteins are subjected to further inspection as described in the text. C: Result of the inspection of the 80 proteins. Proteins are ranked according to the AUC values returned by the arbitrary docking using 25 shortest probes and 10 docking models. Black squares: proteins for which the automatic inspection reveals no multiple interfaces. Red circles: proteins for which the automatic inspection reveals multiple interfaces. Filled circles: visual inspection reveals that arbitrary docking targets a region involved in other interfaces, as exemplified in panel A, open circles imply that visual inspection does not reveal such trend. The vertical red line represents the global AUC obtained for the target data set.