Additional Supporting Information may be found in the online version of this article.

PROT_23182_sm_suppinfofig1.pdf29KSupporting Information Figure S1: Cumulative average target difficulty over all CASPs. The relative difficulty of each CASP is expressed as a Z-score derived from the average similarity of best templates to those target structures and sequences.(see Methods). The higher the score, the more difficult the target set in a specific CASP. This alternative view of target difficulty confirms that CASP9 human/server targets are harder than those of CASP7 and CASP8, and approximately the same in difficulty as CASP5 and CASP6 targets.
PROT_23182_sm_suppinfofig2.pdf194KSupporting Information Figure S2: Trend lines for the average GDT_TS score over the best six groups on each target in all CASPs, as a function of target difficulty. The trend lines here are similar to those for the GDT_TS scores using the best model only (Figure 2, main text), showing that best model performance is reasonably representative of the performance of a number of groups.
PROT_23182_sm_suppinfofig3.pdf99KSupporting Information Figure S3: GDT_TS scores of submitted best models for targets in CASPs 8 and 9, as a function of Z-score ranked target difficulty. Each point represents one target. The three targets ranked easiest and three hardest are omitted for clarity. As in Figure 2, the trend lines show a likely increased accuracy of modeling in the middle range of difficulty in CASP9. This plot differs from Figure 2 in using the Z-score ranking of difficulty, and is included to demonstrate that the apparent mid-range improvement is robust with respect to the choice of difficulty scale.
PROT_23182_sm_suppinfofig4.pdf173KSupporting Information Figure S4: Average, maximum, and minimum spread among the sets of five models submitted for each target, for those groups included in the analysis of best model selection (Figure 8).

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.