Properties of the different protein decoy sets used in this study.

aTraining set (Titan HRD) and test set (Titan HRD*) from the Titan High resolution decoy set [20], available at http://titan.princeton.edu/2010-10-11/Decoys/.bTasser Set II is a structurally non-redundant set of protein structures and decoys derived with the program TASSER. It is available at http://zhanglab.ccmb.med.umich.edu/decoys/.cDecoy sets from the Decoys ‘R’ us repository http://dd.compbio.washington.edu.dDifferent decoy Rosetta-based decoy sets (see text for details), available at http://depts.washington.edu/bakerpg/decoys/.eCollection of models from the successive CASP5 to CASP9 experiments, available from the CASP web site http://predictioncenter.org. CASP-HRD is a high resolution subset of the union of the five sets CASP5 to CASP9, which includes models that have a TM-score larger than 0.5 and a RMSD less than 4 Å to the native structures.fThe Stage_1 and Stage_2 decoy sets used in the CASP10 quality assessment category, available from the CASP web site http://predictioncenter.org. For details on how these sets are prepared, see [48].gAll high and low resolution targets (TSA TM-score>0.5)/(TSA TM-score Files S1 and S2 respectively found in the supporting information.hNprot is the number of different proteins in the dataset, Nres is the average number of residues computed over all proteins in a dataset, and Ndecoys is the average number of decoys per proteins, averaged over the dataset. RMSD, MT, GDT-TS, and Q are the distance measures between the decoys and the corresponding native structures, averaged over all decoys and all proteins. We provide both the average values and the average mean absolute deviations (in parenthesis).Properties of the different protein decoy sets used in this study.