(a) Improvement for all predictions and percent of proteins with at least one correct is plotted for: no filtering (blue), solvent accessibility filtering (green) and residue identity filtering (red). The data presented is from the SCOP superfamily parsed dataset. Symbols are the same as in Fig. 3. In (b), the combination (accessibility + residue identity) filter (magenta) is shown to perform very similar to the residue identity filter alone.