Supplemental Figure 1: Performance of Classification by Entropy Change. A) the Cumulative

Supplemental Figure 1: Performance of classification by entropy change. A) The cumulative distribution function (CDF) of the rank of the lowest RMSD decoy across CASP9 targets is plotted for eight degrees of local character (LC). The LC index is conceptually similar to the metric used in [51, 52], see text for details. The line of equality is shown in black. The LC4 index (yellow) results in the most robust classification when only the change in entropy upon mode excitation is used. B) Classification rank is displayed as a bar plot for LC and other metrics. The percent of targets for which the lowest RMSD decoy appears in the top 10, 50, or 100 is shown. From left to right, the metrics are: LC4, the number of residues with ΔS>0, ΔS<0, standard deviation of ΔS, ΔS=0, Γα, Γβ. It was observed that the distribution of ΔS values for each decoy was well fit to a Gamma distribution (data not shown). Γα and Γβ are the parameters of a Gamma distribution fit to the distribution of ΔS values for each decoy. C) Representative of ΔS for T0516 tetrameric decoys. Each of the 16 decoys is colored separately showing that the structural differences present do result changes to the computed entropy change.

Supplemental Figure 2: Result of classifying CASP9 monomer decoys using entropy, energy, and a combined methodology. Colored bars indicate the range of RMSD100 as defined in [53] and the 20th percentile of RMSD100. Lines show the index of the best monomer classification of CASP9 targets using the indicated formula. We find that the energetic contributionto ΔG is a more powerful independent classifier compared to the entropic contribution as calculated from the ENMs. However, the inclusion of ΔS does improve the classification (see text for details). More rigorous assessment of ΔS will likely lead to further improvementsin the classification scheme. The inclusion of full-structure effects upon this classification, as performed here, represents a step forward in the theory of protein structure evaluation.

Supplemental Figure 3: Distribution of minimum RMSD decoys across the CASP targets. For all 111 CASP9 targets used, we report the RMSD and RMSD100 of the minimum RMSD decoy after performing a sequence alignment to the experimentally determined structure (native) fitting equivalent residues. This step is necessary since many structures have unresolved loops, His-tags, etc.

Supplemental Figure 4:RMSD of the “best” decoy for each CASP targets sorted independently by ΔS (black) or ΔE (red). The true lowest RMSD decoy is plotted with the blue line. Almost all targets have predicted structures within 3Å RMSD of the native. The largest challenge is determining which of the hundreds of decoys is the most native-like. Our 4-body potential is the best at this determination, but still has much room for improvement.