Table 1.

Summary of refinement results for three models with synthetic density maps

Protein Data Bank ID code	Initial rmsd	5 Å						10 Å
		Optimum restraints			No restraints			Optimum restraints			No restraints
		rmsd	CWork	CFree	rmsd	CWork	CFree	rmsd	CWork	CFree	rmsd	CWork	CFree
1AKE	3.60	1.40	0.874	0.211	1.86	0.880	0.189	1.47	0.907	0.524	2.67	0.911	0.425
1IKN	7.73	1.80	0.822	0.355	7.94	0.853	0.175	2.14	0.859	0.441	8.30	0.871	0.354
1HRD	5.96	3.88	0.832	0.235	3.99	0.850	0.167	4.30	0.854	0.365	4.95	0.874	0.316

The three models were taken from the homology model benchmark set from the work by Topf et al. (8) and represent an easy (1AKE), an intermediate (1IKN), and a hard (1HRD) case in terms of structural similarity between starting and target models. Refinements were done with and without DEN restraints for two resolutions: 5 and 10 Å. Results for the optimum restraints correspond to the DEN parameters that lead to the highest free density map correlation, C_free. The rmsd (Å) of the refined to the target structure is always lower when using optimal restraints. The work map correlation, C_work, is always higher without restraints compared with using optimum restraints, because the model fits the density map much better. However, without restraints, the rmsd of the refined to the target structure is always higher, indicating that the density is overfitted. C_free is always higher when refining with optimum restraints compared with refinements without restraints, and higher C_free values always correspond to better structures with lower rmsd.