Figure 3
From: Machine learning to estimate the local quality of protein crystal structures
bCC and RSCC. (a) Arg67 of 3F9X (molecule D) in the correct and incorrect model structures. The template of homology modeling for the incorrect structure was 1ZKK. Electron densities used for the calculation of bCC (left) and RSCC values (right) are also shown. Values of bCC include the environment surrounding Arg67 (left), while RSCC measured correlation electron density covering Arg67 and its atoms (right). (b) Comparison of bCC and RSCC in the region where the main chain is completely misplaced from the correct electron density map. The bCC (left) and RSCC (right) values at residues 7–11 of the model structure at 1.25 Å- and 3.0 Å-resolution structures of 3F9X. The template of homology modeling for the incorrect structure was 5V2N. All the Cα atoms of the region are more than 1 Å away from the corresponding Cα atom of the correct structure. Because bCC is not affected by model bias, no residue is correlated with the correct electron density (bCC < 0.3). Contrastingly, because RSCC is affected by model bias and large B-factors, it shows correlation with electron density, and the effect is larger at 3.0 Å resolution. The electron density maps and coordinates are shown in Supplementary Fig. S1.
