"... In PAGE 12: ... We have therefore also examined how well the different energy functions discriminate native-like from non native-like conformations. Table8 shows the rank scores, Qf scores and Qp scores for the best-scoring native-like conformations. Here we find that of the different proteins in the test set, the native-like conformations of 4rxn are most easily found, and those of 1sn3 least easily.... ..."

"... In PAGE 13: ... In general the Z-score will increase if more of the decoys are native-like thus making the direct comparison of Z-scores uninformative. Table9 shows a more useful Z-score measure for native-like conformations. Here the Z-scores are calculated as averages over all native-like conformations for each protein-energy function combination.... ..."

"... In PAGE 14: ... combinations have significantly greater discriminat- ing power. The combined potential functions are also improved in their ability to pick out native-like conformations ( Table12 ). Parallel with the improve- ments in X-ray conformation identification we see that the Surface + VdW(HL), VdW(MJ) + VdW(HL) and VdW(MJ) + Histogram combinations are better than their constituent functions, but by less than for X-ray conformation identification.... ..."

"... In PAGE 14: ... They mostly parallel the results in Table 11 for Q-scores; we provide them for comparison with other studies. The average Z-scores for all native-like confor- mations, shown in Table14 , parallel those from X-ray Z-scores and X-ray and native-like Q-scores. Once again we see that the Z-scores do not differentiate one combination from another well.... ..."

"... In PAGE 12: ... Three of the functions are able to rank the X-ray structure first for individual proteins but no energy function can consistently rank X-ray structures highly for all proteins. However, some trends in the data of Table6 are discernible. To see these more easily we calculated the quality factor, Q(r) for each rank (see Methods), and then averaged these values Q(r) over different proteins for the same function ( Qf ) and over different functions for the same protein ( Qp ).... In PAGE 12: ... This discrepancy is an indication that our ensembles are a more challenging test of certain energy functions. The Z-scores averaged over different proteins also show that the energy functions, VdW(MJ), VdW(HL) and Histogram, do better than the three others, paralleling what is seen in Table6 using Q-scores. Finding near-native folds As important a characteristic of a potential function as being able to identify the X-ray conformation from a group of incorrect confor- mations is, it is perhaps more important for a potential function to be able to identify native-like conformations.... ..."

"... In PAGE 4: ...4 were derived on 5 out of 18 proteins used as the training set and transferred without further changes to the remaining 13 proteins used as the testing set (see Experimental Procedures). Native-like structures are observed in the course of REMC folding simulations for most proteins regardless of their class ( Table1 ). However, generation of such structures is not sufficient- an objective criterion for structure selection that does not rely on the known native structure needs to be employed.... In PAGE 6: ... For 1CSP, two strands in the N terminus have different alignment from the native, resulting in high RMSD value (Figure 4C). In both cases of 1SHFA and 1CSP, the control runs starting from the native state show lower RMSD and energy than runs starting from random coils and the gap in energies between runs starting from native and random coil structures is much larger than that for other proteins ( Table1 ). This means that the failure for these two proteins to reach a more native-like state may be due to conformational sampling.... ..."

"... In PAGE 13: ... Our most important result is that, due to the small number of points selected and the coarse rotameric de- scription, computing times with Gaussian mapping are highly reduced, while preserving sampling of native- like poses. In Table4 we compare the computing times and resolutions using the standard spheres gen- erated with SPHGEN and the interaction points gener- ated with GAGA for two representative cases. These two cases were selected because they are provided as examples within the different DOCK distributions.... In PAGE 13: ... The total docking time with DOCK 3.0 using our Gaussians and coarse rotamer de- scription in a R12000, with 2% native-like poses in the upper ranking list, is 10 s ( Table4 ). While conditions... In PAGE 14: ... Clustering of conformers [40], or the use of simplified representations for the ligand, can potentially reduce these docking times even further. The data presented in Table4 also suggest that, in practical applications, additional filtering might be required. Re-ranking the upper-ranking configurations selected with the CONTACT scoring using more soph- isticated energy functions, such as the GB-SA method [41], might be of interest, and will be the subject of future research.... ..."

"... In PAGE 6: ...7. Prediction of N-mer assemblies Table3 lists the name of the PDB code of the com- plex crystal structure, the rank of the predictions with of the of 238 S.... In PAGE 9: ...poorly for the membrane protein 1P7B, where we again failed to retain any low free energy conformation (see Table3 ). For 1BQ4 and 1YKF, we actually found the dimers, however, the D2 assembly had clashes that failed the overlap criterion and therefore the native-like clus- ters got discarded.... In PAGE 10: ...e., D2,DC3,andC3D, the cluster size of each interface listed in Table3 provides an indication of which inter- face forms first. This is consistent with the underlying principle that cluster size correlates with the lowest free energy minimum, the larger the cluster size of one inter- face is the more likely that it forms first.... ..."

"... In PAGE 15: ... All participants judged 79% or above (range = 79% - 100%) of the critical items correctly in this task, suggesting that they had acquired near-native or native-like knowledge in this area of English grammar. Table1 summarises the results from the two pretests. Table 1: Mean proficiency and grammaticality judgement scores from the German L2 learners... ..."