Limits...
Local error estimates dramatically improve the utility of homology models for solving crystal structures by molecular replacement.

Bunkóczi G, Wallner B, Read RJ - Structure (2015)

Bottom Line: Predicted structures submitted for CASP10 have been evaluated as molecular replacement models against the corresponding sets of structure factor amplitudes.It has been found that the log-likelihood gain score computed for each prediction correlates well with common structure quality indicators but is more sensitive when the accuracy of the models is high.In addition, it was observed that using coordinate error estimates submitted by predictors to weight the model can improve its utility in molecular replacement dramatically, and several groups have been identified who reliably provide accurate error estimates that could be used to extend the application of molecular replacement for low-homology cases.

View Article: PubMed Central - PubMed

Affiliation: Department of Haematology, Cambridge Institute for Medical Research, University of Cambridge, Hills Road, Cambridge CB2 0XY, UK.

Show MeSH

Related in: MedlinePlus

Typical LLG versus GDT_TS Scatter Plots Observed for Targets(A) Target TR705 contains two domains and refinement of one of these was requested. If the second domain is not taken into account in the likelihood calculations, the black curve is obtained, which shows no correlation between the two scores. However, by taking the contribution from the second domain into account (grey curve), a clear correlation is obtained (for scores shown, the contribution of the second domain alone is subtracted for the plot). ASU, asymmetric unit.(B) Uninformative LLG plot for target T0653 with all models falling into the low accuracy zone.(C) Very sensitive LLG plot for target T0717, domain 2 (taking the unpredicted domain 1 into account). Predictors have managed to model residues Val67 to Gly119 (out of 166 residues) very accurately, and this gives a clear signal in scoring with the 1.9 Å X-ray data. For the “outlier” models above GDT_TS = 35, the accuracy of the named residue segment is comparable with that of the rest of the structure.(D) Atypically small signal observed for target T0704.
© Copyright Policy - CC BY
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4321884&req=5

fig1: Typical LLG versus GDT_TS Scatter Plots Observed for Targets(A) Target TR705 contains two domains and refinement of one of these was requested. If the second domain is not taken into account in the likelihood calculations, the black curve is obtained, which shows no correlation between the two scores. However, by taking the contribution from the second domain into account (grey curve), a clear correlation is obtained (for scores shown, the contribution of the second domain alone is subtracted for the plot). ASU, asymmetric unit.(B) Uninformative LLG plot for target T0653 with all models falling into the low accuracy zone.(C) Very sensitive LLG plot for target T0717, domain 2 (taking the unpredicted domain 1 into account). Predictors have managed to model residues Val67 to Gly119 (out of 166 residues) very accurately, and this gives a clear signal in scoring with the 1.9 Å X-ray data. For the “outlier” models above GDT_TS = 35, the accuracy of the named residue segment is comparable with that of the rest of the structure.(D) Atypically small signal observed for target T0704.

Mentions: LLG scores were calculated for all predictions submitted for CASP10 targets that were determined by X-ray crystallography and for which the measured X-ray diffraction amplitudes were available. The relationship with the GDT_TS score was examined on scatter plots (Figure 1). For the majority of cases, the LLG score showed a clear functional relationship with the GDT_TS score. At GDT_TS < 40–50, the LLG scores were almost constant (average Spearman correlation coefficient for segment GDT_TS ≤ 50, 36%); they started to increase slowly for GDT_TS values above 50–60 and then very rapidly for GDT_TS > 80 (average Spearman correlation coefficient for segment GDT_TS > 50, 71%), although some deviations have also been observed (Figures 1C and 1D). This suggests that the LLG score cannot discriminate among predictions with large errors but can accurately rank good-quality predictions.


Local error estimates dramatically improve the utility of homology models for solving crystal structures by molecular replacement.

Bunkóczi G, Wallner B, Read RJ - Structure (2015)

Typical LLG versus GDT_TS Scatter Plots Observed for Targets(A) Target TR705 contains two domains and refinement of one of these was requested. If the second domain is not taken into account in the likelihood calculations, the black curve is obtained, which shows no correlation between the two scores. However, by taking the contribution from the second domain into account (grey curve), a clear correlation is obtained (for scores shown, the contribution of the second domain alone is subtracted for the plot). ASU, asymmetric unit.(B) Uninformative LLG plot for target T0653 with all models falling into the low accuracy zone.(C) Very sensitive LLG plot for target T0717, domain 2 (taking the unpredicted domain 1 into account). Predictors have managed to model residues Val67 to Gly119 (out of 166 residues) very accurately, and this gives a clear signal in scoring with the 1.9 Å X-ray data. For the “outlier” models above GDT_TS = 35, the accuracy of the named residue segment is comparable with that of the rest of the structure.(D) Atypically small signal observed for target T0704.
© Copyright Policy - CC BY
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4321884&req=5

fig1: Typical LLG versus GDT_TS Scatter Plots Observed for Targets(A) Target TR705 contains two domains and refinement of one of these was requested. If the second domain is not taken into account in the likelihood calculations, the black curve is obtained, which shows no correlation between the two scores. However, by taking the contribution from the second domain into account (grey curve), a clear correlation is obtained (for scores shown, the contribution of the second domain alone is subtracted for the plot). ASU, asymmetric unit.(B) Uninformative LLG plot for target T0653 with all models falling into the low accuracy zone.(C) Very sensitive LLG plot for target T0717, domain 2 (taking the unpredicted domain 1 into account). Predictors have managed to model residues Val67 to Gly119 (out of 166 residues) very accurately, and this gives a clear signal in scoring with the 1.9 Å X-ray data. For the “outlier” models above GDT_TS = 35, the accuracy of the named residue segment is comparable with that of the rest of the structure.(D) Atypically small signal observed for target T0704.
Mentions: LLG scores were calculated for all predictions submitted for CASP10 targets that were determined by X-ray crystallography and for which the measured X-ray diffraction amplitudes were available. The relationship with the GDT_TS score was examined on scatter plots (Figure 1). For the majority of cases, the LLG score showed a clear functional relationship with the GDT_TS score. At GDT_TS < 40–50, the LLG scores were almost constant (average Spearman correlation coefficient for segment GDT_TS ≤ 50, 36%); they started to increase slowly for GDT_TS values above 50–60 and then very rapidly for GDT_TS > 80 (average Spearman correlation coefficient for segment GDT_TS > 50, 71%), although some deviations have also been observed (Figures 1C and 1D). This suggests that the LLG score cannot discriminate among predictions with large errors but can accurately rank good-quality predictions.

Bottom Line: Predicted structures submitted for CASP10 have been evaluated as molecular replacement models against the corresponding sets of structure factor amplitudes.It has been found that the log-likelihood gain score computed for each prediction correlates well with common structure quality indicators but is more sensitive when the accuracy of the models is high.In addition, it was observed that using coordinate error estimates submitted by predictors to weight the model can improve its utility in molecular replacement dramatically, and several groups have been identified who reliably provide accurate error estimates that could be used to extend the application of molecular replacement for low-homology cases.

View Article: PubMed Central - PubMed

Affiliation: Department of Haematology, Cambridge Institute for Medical Research, University of Cambridge, Hills Road, Cambridge CB2 0XY, UK.

Show MeSH
Related in: MedlinePlus