Application of Protein–Ligand NOE Matching to the Rapid Evaluation of Fragment Binding Poses - Fragment-Based Drug Discovery - page 105

The three examples described in this section all involve relatively small, well-behaved
proteins in complex with lead-like compounds. The results obtained indicate that, for lead-
like or drug-like compounds bound to suitable targets, NOE matching can yield relatively
accurate binding poses without using any protein NMR assignments or accurate chemical
shift predictions. Later in this chapter, we will focus on challenges presented by simpler
compounds and/or large proteins for NOE matching in particular and for detecting and
analyzing protein-ligand NOE interactions in general.
5.4 Enhanced Pose Generation and Pose Scoring
For NOE matching to identify the correct pose accurately, it is essential that the ensemble
of poses contains one or more poses that are very similar to the true pose. Generation of
a broad sampling of poses with existing software, however, is not straightforward. Most
pose generation algorithms, such as Glide, [ 23 ] are designed to identify the optimal pose,
discarding false poses. Identification of the true pose by NOE matching, on the other hand,
relies on the fact the COST of correct poses should be significantly lower than the COST
of decoy poses. Therefore, it is important that a wide distribution of poses is generated and
evaluated, as this increases the probability that the lowest COST poses obtained during an
NOEmatching run reflect a global minimumof the COST function and not a local minimum.
To address this concern, we have adapted an internally developed posing engine, Poser . For
predefined ligand and protein conformations, Poser provides a systematic and exhaustive
sampling of poses in a binding site.
Poser generates all possible poses of user-supplied ligand conformations within a spe-
cified resolution. The only limitation imposed is the absence of significant steric clashes
between the ligand and protein. A regularly spaced grid is centered on a user-specified
binding site. By default, the binding site consists of the entire protein, but can be specified
by defining the coordinate boundaries of the binding site.Amask is created by labeling each
grid point as either inside the protein or outside the protein. Further, the shortest distance
of each grid point to the protein surface is also calculated. For these purposes, the atomic
radii for the protein are set to 90% of the Bondi atomic radii, as specified in OEChem,
v1.4. [ 24 ] This softening of the protein radii is performed to account for small displacements
that may occur, but that are not explicitly modeled.
The geometric centroid of each conformation of each ligand is placed iteratively on every
grid point that is not labeled as being inside the protein. The molecule is then rotated about
the grid point by a user-defined number of degrees. Each resulting pose is then checked
for bumps by making sure the pose does not overlap a grid point labeled as being inside
the protein. The user may specify a scaling factor to decrease the Bondi radii of the ligand
atoms for these purposes. If the pose does not exceed the user-specified limit of allowable
bumps, the pose is written out and saved.
If the user requires contact between the ligand and the protein, the maximum distance
from the geometric centroid of each conformation to its molecular surface is calculated. The
molecule is only placed on those grid points that are labeled as being within this distance
plus 1 Å.
Examples of samplings obtained by Poser are provided in Table 5.1. To generate the
input conformers for Poser , we used the program Omega [ 25 ] with standard defaults. This