Fig. 1
Flowchart of the across-proteome modeling of dimer structures and the prediction of protein-protein interactions. a Query protein structures are first built with homology modeling. b Subsequently, a binding site is identified in the receptor and initial dimer models are generated through rigid body docking. c Initial models are then re-ranked by eRankPPI taking into account the binding site information and (d) subjected to a flexible refinement. e Machine learning followed by (f) annotation filters are finally employed to identify biologically relevant protein assemblies (g)