An alternative approach is to build loop models ab initio. The
program Rosetta offers multiple algorithms for this purpose.
Including a Monte Carlo Cyclic coordinate descent method
[56, 57], and Kinematic closure methods [58] that have been
implemented for both ab initio loop modeling and loop refinement. In spite of these advances, if the resolution is not sufficiently
high, it may be “safer” to generate multiple loop conformations
and cluster them to provide a better representation of the fit to the
density, as done in the case of MKLP2 described above [22] (see
Subheading 3.4).
3.6.2 Ligand Fitting
At near-atomic resolutions, where the density allows for a more
accurate placement of atoms, it is sometimes possible to refine
ligand directly in the density. This was the case with the ligand
cadazolid an inhibitor of the bacterial 50S ribosomal subunit. The
authors used real-space refinement in PHENIX [18] to model the
ligand directly into the density in a 3 A ˚ map (PDB ID: 6QUL,
EMDB: 4638) [59]. However, for lower resolution maps the correct placement of ligand atoms into the density can be ambiguous.
One common approach that can aid this process is molecular
(or ligand) docking. This involves the prediction of the predominant binding mode(s) of a ligand with a protein, given its threedimensional structure. Multiple conformations of the complex are
generated, scored, and ranked. Scoring functions for docking programs aim to estimate the free energy of the complex, with more
stable conformations ranked higher than others, which has the
advantage of considering the effects of physical interactions
between the ligand and protein. However, the results can be heavily
biased by the accuracy of the positions of the side chains that line
the binding sites, which for intermediate resolution cryo-EM maps
is not adequate. This problem can be somewhat attenuated by
programs such as GOLD [60] or HADDOCK [61] that allow for
side-chain flexibility during docking calculations. Additionally, the
ability of docking software to generate a correct ligand conformation is not ideal [62]. Multiple studies have shown that the confidence in identifying correct conformations can be increased by
consensus docking [63]. Where the ligand is docked into the
receptor using multiple docking software and conformations predicted by more than one program are taken to be correct.
A difference map is the subtraction of one density map from
another and represents a way of identifying changes in conformations and compositions of the complex. Depending on the resolution, it is sometimes possible to identify density corresponding to a
bound ligand if one map contains the ligand while the other represents an unbound state. This operation is implemented in many
software packages that allow for the analysis of 3D cryo-EM maps,
such as Chimera [25] or TEMPy [17]. However, for a difference to
contain meaningful information certain criteria must be met,
206
Tristan Cragnolini et al.
program Rosetta offers multiple algorithms for this purpose.
Including a Monte Carlo Cyclic coordinate descent method
[56, 57], and Kinematic closure methods [58] that have been
implemented for both ab initio loop modeling and loop refinement. In spite of these advances, if the resolution is not sufficiently
high, it may be “safer” to generate multiple loop conformations
and cluster them to provide a better representation of the fit to the
density, as done in the case of MKLP2 described above [22] (see
Subheading 3.4).
3.6.2 Ligand Fitting
At near-atomic resolutions, where the density allows for a more
accurate placement of atoms, it is sometimes possible to refine
ligand directly in the density. This was the case with the ligand
cadazolid an inhibitor of the bacterial 50S ribosomal subunit. The
authors used real-space refinement in PHENIX [18] to model the
ligand directly into the density in a 3 A ˚ map (PDB ID: 6QUL,
EMDB: 4638) [59]. However, for lower resolution maps the correct placement of ligand atoms into the density can be ambiguous.
One common approach that can aid this process is molecular
(or ligand) docking. This involves the prediction of the predominant binding mode(s) of a ligand with a protein, given its threedimensional structure. Multiple conformations of the complex are
generated, scored, and ranked. Scoring functions for docking programs aim to estimate the free energy of the complex, with more
stable conformations ranked higher than others, which has the
advantage of considering the effects of physical interactions
between the ligand and protein. However, the results can be heavily
biased by the accuracy of the positions of the side chains that line
the binding sites, which for intermediate resolution cryo-EM maps
is not adequate. This problem can be somewhat attenuated by
programs such as GOLD [60] or HADDOCK [61] that allow for
side-chain flexibility during docking calculations. Additionally, the
ability of docking software to generate a correct ligand conformation is not ideal [62]. Multiple studies have shown that the confidence in identifying correct conformations can be increased by
consensus docking [63]. Where the ligand is docked into the
receptor using multiple docking software and conformations predicted by more than one program are taken to be correct.
A difference map is the subtraction of one density map from
another and represents a way of identifying changes in conformations and compositions of the complex. Depending on the resolution, it is sometimes possible to identify density corresponding to a
bound ligand if one map contains the ligand while the other represents an unbound state. This operation is implemented in many
software packages that allow for the analysis of 3D cryo-EM maps,
such as Chimera [25] or TEMPy [17]. However, for a difference to
contain meaningful information certain criteria must be met,
206
Tristan Cragnolini et al.