Median Performance Across the 12-Residue Benchmark Sets and Illustration of Synergy.

<p>(A) Barplot showing median percent sub-Angstrom (m%sA) across benchmark sets 1 and 2 for the individual sampling strategies tested here as well as their combination (“all five”), showing a clearly increased percentage of sub-Angstrom models. Error bars are standard deviations from 3 independent simulations (generating 500 models for each of the cases in the dataset, repeated three times). Colors are as in <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0063090#pone-0063090-g002" target="_blank">Figure 2</a>. The value for CCD is 0 with this measure. (B) Barplots of “leave-one-out” trials in which all combinations of 4 sampling improvement methods are tested: without Taboo sampling (red, NGK), without Omega sampling (dark green), without Rama2b (green), without Ramp repulsive (dark yellow), and without Ramp rama (dark orange). These data are also provided in <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0063090#pone-0063090-t001" target="_blank">Table 1</a>. (C–F) Energy-vs-RMSD plots of remodeling PDB 1oyc with standard KIC (C), next-generation KIC (D), Rama2b sampling (E) and Ramp repulsive sampling (F). REU, Rosetta-energy units. (G) RMSD distributions for the different methods. Colors are as in <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0063090#pone-0063090-g002" target="_blank">Figure 2</a>, NGK in red. Rama2b and Ramp repulsive sampling both contribute to enabling sampling of sub-Angstrom conformations of the remodeled segment (in 1oyc.pdb), while the other individual strategies do not change the RMSD distribution for this case. Nevertheless, the combined performance in NGK is higher than expected from the individual improvements, indicating synergy. (H) Boxplots of median RMSDs for standard KIC (blue), CCD (gray) and NGK (red), based on the lowest-energy model for each benchmark case. Boxplots show the minimum and maximum among the lowest-scoring RMSDs across the benchmark set (error bars), the 25^th and 75^th percentile (box boundaries) as well as the median (thick line). Both kinematic-closure-based methods have lower median RMSDs than CCD. (I) Comparison of the lowest-scoring RMSD for each benchmark case in simulations with standard KIC vs. with NGK. NGK typically achieves lower RMSDs than standard KIC (red), while for some cases KIC achieves lower RMSDs (blue). Cases with an RMSD change <10% are shown in black.