2.3.2 Local Scoring
Although it is sometimes convenient to provide a single number
summarizing the quality of a fit, for example during an optimization method such as a rigid-body fit, it is also useful to have a more
local information regarding the quality of a fit: a loop or some other
structural elements may not fit well to the map, if for example there
is some conformational difference between the proposed structure
and the one represented by the map. This can be assessed by using a
local score that will provide several numerical values, usually along
voxels or groups of voxels and this can be guided by the sequence or
known structural elements (such as secondary structure or
domains)
Local
Cross-Correlation: SCCC
The segment-based cross-correlation score (SCCC) is computed on a
subset of the map, using the cross-correlation (see Note 2). The
part of the map that is chosen for each SCCC calculation is based on
a segmentation of the protein into rigid bodies, or along its
sequence. The initial identification of segment (rigid bodies) in
the structure to be assessed can be done using RIBFIND (see
Subheading 3.2). After normalization of the data, and alignment
of the fragments, the score is calculated.
SMOC
The SMOC score is computed in the same way, using the Manders’
coefficient presented above. The score can be calculated over overlapping windows of neighboring residues along the sequence
(SMOCf), or by selecting voxels in the map close to the residues’
atoms (SMOCd, see [15]), and computing the Manders’ overlap
coefficient over these voxels. By repeating this procedure over
windows of nine residues at a time, moving the window by one
residue each time, it is possible to obtain an SMOCf profile over the
entire protein sequence (and structure), giving an indication as to
which regions are well fitted in the map, and which are not [16].
Several of those functions have been developed and tested in
our group [10, 11]. The overlap score displayed excellent properties
in terms of discriminating good and bad fits. A combination of
OVR and SMOC had better statistical properties for low-resolution
maps, with all scores displaying similar performances at high
resolution.
2.3.3 Consensus Scoring
The different scoring methods are not always consistent with one
another, with different models getting ranked lower or higher. This
is problematic when limitations in the processing require the choice
of a limited subsets, or even a single model to use in further
refinement. By computing the score and ranks with multiple methods, and combining them using the Borda score, it has been shown
that most of the ambiguity in the rankings can be resolved [17].
All the scores described above are available within TEMPy
[17]. TEMPy is a python library developed for the assessment
and manipulation of macromolecular structures specifically
CryoEM Density Fitting and Validation
193
Although it is sometimes convenient to provide a single number
summarizing the quality of a fit, for example during an optimization method such as a rigid-body fit, it is also useful to have a more
local information regarding the quality of a fit: a loop or some other
structural elements may not fit well to the map, if for example there
is some conformational difference between the proposed structure
and the one represented by the map. This can be assessed by using a
local score that will provide several numerical values, usually along
voxels or groups of voxels and this can be guided by the sequence or
known structural elements (such as secondary structure or
domains)
Local
Cross-Correlation: SCCC
The segment-based cross-correlation score (SCCC) is computed on a
subset of the map, using the cross-correlation (see Note 2). The
part of the map that is chosen for each SCCC calculation is based on
a segmentation of the protein into rigid bodies, or along its
sequence. The initial identification of segment (rigid bodies) in
the structure to be assessed can be done using RIBFIND (see
Subheading 3.2). After normalization of the data, and alignment
of the fragments, the score is calculated.
SMOC
The SMOC score is computed in the same way, using the Manders’
coefficient presented above. The score can be calculated over overlapping windows of neighboring residues along the sequence
(SMOCf), or by selecting voxels in the map close to the residues’
atoms (SMOCd, see [15]), and computing the Manders’ overlap
coefficient over these voxels. By repeating this procedure over
windows of nine residues at a time, moving the window by one
residue each time, it is possible to obtain an SMOCf profile over the
entire protein sequence (and structure), giving an indication as to
which regions are well fitted in the map, and which are not [16].
Several of those functions have been developed and tested in
our group [10, 11]. The overlap score displayed excellent properties
in terms of discriminating good and bad fits. A combination of
OVR and SMOC had better statistical properties for low-resolution
maps, with all scores displaying similar performances at high
resolution.
2.3.3 Consensus Scoring
The different scoring methods are not always consistent with one
another, with different models getting ranked lower or higher. This
is problematic when limitations in the processing require the choice
of a limited subsets, or even a single model to use in further
refinement. By computing the score and ranks with multiple methods, and combining them using the Borda score, it has been shown
that most of the ambiguity in the rankings can be resolved [17].
All the scores described above are available within TEMPy
[17]. TEMPy is a python library developed for the assessment
and manipulation of macromolecular structures specifically
CryoEM Density Fitting and Validation
193