Examples illustrating optimal mixing and matching of the best predicted secondary structures. A. Comparative models of CASP7 target T0369 (PDB identifier 2H4O) are shown to explain the process of mixing and matching between two initial models. The backbone ribbon representations of the experimental structure and the two initial models are indicated as grey, green and red, respectively. Helix 1 of initial model 1 (red) is not accurately modeled, but helix 3 is predicted reasonably, resembling the experimental conformation most. Helix 1 of initial model 2 (green) is properly modeled, but helix 3 is not predicted in correct conformation. The CF method incorporates helix 3 from initial model 1 and helix 1 from initial model 2, resulting in a CF model that is improved by 0.4 Å CαRMSD relative to the experimental structure. (Figures were prepared with Molscript  and Raster3D .). B. Comparative models of CASP7 target T0332 (PDB identifier 2HA8) are shown to explain the process of mixing and matching between multiple initial models. The backbone ribbon representations of the experimental structure (grey) is shown compared to the refined CF model. The CF method assembles β-strand 3 from initial model 1. (green), helix 2 from initial model 2 (red) and a major loop at the bottom of the central β-sheet from initial model 3 (yellow), which are the best predicted substructures in the three initial models, respectively. This produces a CF model that is improved by 0.8 Å CαRMSD relative to the experimental structure.