Contact prediction in protein modeling: Scoring, folding and refinement of coarse-grained models

BMC Structural Biology

Table 3 Comparison of different approaches to contact-based modeling tested in this work.

Set of contact data^(a)	Number of predicted contacts		avg. Cα RMSD [Å]
		Contact-based ranking	De novo folding		Refinement
			First ^(c)	Best ^(d)	First ^(c)	Best ^(d)
N/2 top-scoring contacts from each of the best two predictors	N	9.58	8.93	7.53	7.69	7.10
N/2 top-scoring contacts from each of the best three predictors	1.5 N	9.96	8.69	8.02	8.23	7.36
Consensus of the whole data from the best three predictors	N/2 N 1.5 N	9.82 9.80 9.81	8.15 7.35 8.03	7.14 8.11 8.68	8.28 8.11 8.68	7.17 7.30 7.73
Consensus of the whole data from the best five predictors	N/2 N 1.5 N	9.83 9.91 9.79	8.92 8.94 8.70	7.92 7.65 7.57	8.44 7.89 7.71	7.02 7.21 7.11
Consensus of the whole data from all nine predictors	N/2 N 1.5 N	9.70 9.77 9.64	9.26 8.79 8.42	7.98 7.52 7.07	8.06 7.63 8.02	6.44 6.41 6.99

^(a) The same as in Table 2.
^(b) RMSD computed only for models ranked as first.
^(c) RMSD component computed only for that model of each target which was likely to be selected as the first model (the most probable), because it was a centroid structure of the most populated cluster obtained in a simulation.
^(d) RMSD component computed only for the best model of each target (a centroid structure of a cluster which was most similar to the native structure).
The performance of each method is presented as the Cα RMSD of the final protein model averaged over 14 targets selected, for the sake of shortening computation time, from NF and FR/A categories (i.e. T0198, T0199-3, T0209-1, T0212, T0215, T0230, T0235-2, T0239, T0248-1, T0262-1, T0272-1, T0272-2, T0280-2, T0281). The average Cα RMSD of the first models for this subset of 14 CASP6 targets for the Kolinski-Bujnicki group was 10.27 and we observed improvement of this value for every method and every contact data set. The lowest average RMSD was obtained in the case of the refinement simulations with contact restraints based on the data from the best two predictors.

ISSN: 1472-6807