designed to integrate with cryo-EM density maps. TEMPy implements functions for the reading, writing, assessment, and fitting of
cryo-EM maps and structures within them.
3 Structure-Density Fitting
3.1 Protein Structure
Prediction
The specific approach used for generating an atomic model from a
cryo-EM map is dependent upon the information available. When
the data contained within the cryo-EM map is at a sufficiently high
resolution that the positions of amino acid side-chain atoms can be
determined, it is possible to build a model de novo into the map
[18, 19]. Although this is becoming more commonplace, in most
cases so far initial candidate models are derived from an a priori
structural model. This can be a known structure of the candidate
protein complex or any of its components from other experiments
(e.g., X-ray crystallography). One issue that can arise in this scenario is incomplete structures (e.g., flexible regions not observed
during experiments). Another issue is differences in conformation
as a result of structure determination at different experimental
conditions. However, it is often the case that a suitable experimental structure is not available at all. In this instance, protein structure
prediction approaches (such as comparative [20, 21] or ab initio
modeling can be used, either to generate models for the missing
components/regions or obtain a complete model of a protein
complex.
Such methodology has been used to generate initial structures
for a kinesin-6 family motor MKLP2 [22]. Briefly, for each nucleotide state of MKLP2 100 homology models were generated using
MODELLERv9.51. Multiple templates were selected using
sequence identity and comparison of the MKLP2 predicted secondary structure with the DSSP-assigned [23] secondary structure
elements of candidate templates. The best models were chosen
using the QMEAN (Qualitative Model Energy ANalysis) score, a
measure of structure quality [24].
3.2 Rigid-Body
Fitting
Once a candidate model is identified, it has to be fitted (docked)
into the EM map. A common approach to this task involves the
systematic maximization of the CCC between the candidate model
and the map. This however is a nontrivial task, especially if the
density corresponding to individual proteins cannot be easily distinguished. Therefore, rigid fitting can be thought of in two parts,
namely global and local rigid fitting. If a reasonable approximation
regarding the initial placement of individual proteins into the map
is possible, a local search to refine the models about their approximate positions may yield a sensible candidate model of the
194
Tristan Cragnolini et al.
cryo-EM maps and structures within them.
3 Structure-Density Fitting
3.1 Protein Structure
Prediction
The specific approach used for generating an atomic model from a
cryo-EM map is dependent upon the information available. When
the data contained within the cryo-EM map is at a sufficiently high
resolution that the positions of amino acid side-chain atoms can be
determined, it is possible to build a model de novo into the map
[18, 19]. Although this is becoming more commonplace, in most
cases so far initial candidate models are derived from an a priori
structural model. This can be a known structure of the candidate
protein complex or any of its components from other experiments
(e.g., X-ray crystallography). One issue that can arise in this scenario is incomplete structures (e.g., flexible regions not observed
during experiments). Another issue is differences in conformation
as a result of structure determination at different experimental
conditions. However, it is often the case that a suitable experimental structure is not available at all. In this instance, protein structure
prediction approaches (such as comparative [20, 21] or ab initio
modeling can be used, either to generate models for the missing
components/regions or obtain a complete model of a protein
complex.
Such methodology has been used to generate initial structures
for a kinesin-6 family motor MKLP2 [22]. Briefly, for each nucleotide state of MKLP2 100 homology models were generated using
MODELLERv9.51. Multiple templates were selected using
sequence identity and comparison of the MKLP2 predicted secondary structure with the DSSP-assigned [23] secondary structure
elements of candidate templates. The best models were chosen
using the QMEAN (Qualitative Model Energy ANalysis) score, a
measure of structure quality [24].
3.2 Rigid-Body
Fitting
Once a candidate model is identified, it has to be fitted (docked)
into the EM map. A common approach to this task involves the
systematic maximization of the CCC between the candidate model
and the map. This however is a nontrivial task, especially if the
density corresponding to individual proteins cannot be easily distinguished. Therefore, rigid fitting can be thought of in two parts,
namely global and local rigid fitting. If a reasonable approximation
regarding the initial placement of individual proteins into the map
is possible, a local search to refine the models about their approximate positions may yield a sensible candidate model of the
194
Tristan Cragnolini et al.