In the case of the lower end of intermediate range and even
lower (worse than 15 A ˚ ), the use of flexible fitting techniques is
limited but may include, for example, refinement of the orientation
between domains.
3.5 Consensus
Refinement Approach
The importance of consensus flexible fitting methods was highlighted in one study that showed that even with diverse methodologies, flexible fitting methods generally converge to the same
solution [52]. This solution was seen to better represent the conformations of the density map when compared to rigid-body fitting
and formed the basis of the idea of using consensus flexible fitting
methodologies to improve the quality of fitted models. The idea is
to compare the results of the different fits produced by different
methods using a local goodness-of-fit score (such as the SCCC
described in Subheading 2.3) [53]. Additionally, the results of all
the fitting programs are compared between themselves, for example, using Cβ RMSD. By this method, consensus local regions
(regions with a low RMSD between programs) can be identified.
Furthermore, non-consensus regions (regions with a high RMSD
between programs) can identify areas where the fit is incorrect and
requires further refinement, and these usually correlate with worse
fit to the density.
This approach has been shown to be able to detect errors
propagated from incorrect comparative models. An actin subunit
comparative model was flexibly fit into a 9 A ˚ resolution simulated
density map from a high-resolution X-ray structure (PDB ID:
2A40, chain A) in a distinct conformation, using both Flex-EM
and iMODFIT [53]. The SCCC scores were calculated from the
SSEs and the average SCCC values were very similar for Flex-EM
and iMODFIT, respectively. The RMSD for the fitted model compared to the ground truth model was seen to be approximately 4 A ˚
for both programs. This apparent inability to converge to the
correct conformation was hypothesized to be due to errors from
the model. Using the QMEAN model assessment score [24], unreliable residues in the initial comparative model were identified.
Identified unreliable residues were seen to be within loop regions
connecting SSEs with low consensus fits. A hybrid approach was
then used by running Flex-EM with the iMODFIT output, and
relaxing constraints on non-consensus SSEs. In this way, the final
model was more representative of the ground truth model (with
averaged Cβ RMSD over all SSEs of 3.6 A ˚ ), with the SCCC
improving in 84% of cases. The protocol was then applied to the
case of the mature and empty capsids of Coxsackievirus A7 (CAV7)
by flexibly fitting comparative models into the corresponding cryoEM density maps at 8.2 and 6.1 A ˚ resolution.
A similar methodology was used more recently to fit comparative models of the mouse MKLP2 (kinesin 6) in complex with ADP.
ALFx, ADP, AMPPNP and in an apo state were determined at 5.5,
202
Tristan Cragnolini et al.
lower (worse than 15 A ˚ ), the use of flexible fitting techniques is
limited but may include, for example, refinement of the orientation
between domains.
3.5 Consensus
Refinement Approach
The importance of consensus flexible fitting methods was highlighted in one study that showed that even with diverse methodologies, flexible fitting methods generally converge to the same
solution [52]. This solution was seen to better represent the conformations of the density map when compared to rigid-body fitting
and formed the basis of the idea of using consensus flexible fitting
methodologies to improve the quality of fitted models. The idea is
to compare the results of the different fits produced by different
methods using a local goodness-of-fit score (such as the SCCC
described in Subheading 2.3) [53]. Additionally, the results of all
the fitting programs are compared between themselves, for example, using Cβ RMSD. By this method, consensus local regions
(regions with a low RMSD between programs) can be identified.
Furthermore, non-consensus regions (regions with a high RMSD
between programs) can identify areas where the fit is incorrect and
requires further refinement, and these usually correlate with worse
fit to the density.
This approach has been shown to be able to detect errors
propagated from incorrect comparative models. An actin subunit
comparative model was flexibly fit into a 9 A ˚ resolution simulated
density map from a high-resolution X-ray structure (PDB ID:
2A40, chain A) in a distinct conformation, using both Flex-EM
and iMODFIT [53]. The SCCC scores were calculated from the
SSEs and the average SCCC values were very similar for Flex-EM
and iMODFIT, respectively. The RMSD for the fitted model compared to the ground truth model was seen to be approximately 4 A ˚
for both programs. This apparent inability to converge to the
correct conformation was hypothesized to be due to errors from
the model. Using the QMEAN model assessment score [24], unreliable residues in the initial comparative model were identified.
Identified unreliable residues were seen to be within loop regions
connecting SSEs with low consensus fits. A hybrid approach was
then used by running Flex-EM with the iMODFIT output, and
relaxing constraints on non-consensus SSEs. In this way, the final
model was more representative of the ground truth model (with
averaged Cβ RMSD over all SSEs of 3.6 A ˚ ), with the SCCC
improving in 84% of cases. The protocol was then applied to the
case of the mature and empty capsids of Coxsackievirus A7 (CAV7)
by flexibly fitting comparative models into the corresponding cryoEM density maps at 8.2 and 6.1 A ˚ resolution.
A similar methodology was used more recently to fit comparative models of the mouse MKLP2 (kinesin 6) in complex with ADP.
ALFx, ADP, AMPPNP and in an apo state were determined at 5.5,
202
Tristan Cragnolini et al.