Application of Protein–Ligand NOE Matching to the Rapid Evaluation of Fragment Binding Poses - Fragment-Based Drug Discovery - page 100

natural products, isotope labeling of the ligand may also be required to obtain bound-state
assignments.) A set of possible binding poses is generated by any suitable method or com-
bination of methods. In general, the sampling of binding poses must be extensive, as poses
that are similar to the true pose must be sampled if they are to be identified by the NOE
matching procedure. Each pose is used to predict a 3D X-filtered NOESY spectrum. The
chemical shifts used for these predicted spectra can be derived from actual experimental
assignments, from chemical shift prediction algorithms (as described later in this chapter)
or most simply by using the average chemical values for diamagnetic proteins available
from the Biological Magnetic Resonance Data Bank (BMRB). [ 21 ]
The central idea of NOE matching is to determine how well the predicted 3D X-filtered
NOESY spectrummatches to the experimental 3D X-filtered NOESY spectra. Poses which
produce a good match between predicted and experimental spectrum are expected to
resemble more closely the true binding pose. This expectation depends critically on having
enough information in the observed pattern of NOEs to define the binding pose in a nonde-
generate fashion. NOE matching casts the problem of measuring the similarity between
predicted and experimental spectra as an equally partitioned bipartite graph weighted
matching problem, [ 22 ] wherein experimentally identified 1 H 13 C groups are matched to HC
atom groups predicted to give rise to NOEs by the given pose. In addition to providing
a COST value for each pose, potential assignments for the experimental NOE peaks are
generated. Ahypothetical example of a 3D X-filtered NOESYmatching problem cast as an
equally partitioned bipartite graph is shown in Figure 5.2. The algorithm used [ 22 ] finds an
Figure 5.2 Equally partitioned bipartite graph representing a hypothetical instance of the 3D
X-filtered NOESY bipartite graph weighted matching problem. Three ligand 1 H resonances and
four protein 1 H 13 C groups are involved in experimentally observed NOEs. The binding pose
predicts NOEs involving the same three ligand 1 H atoms and five protein HC groups. Each
edge between the observed and predicted nodes has an edge weight value associated with it.
The sum of these edge weights defines the COST of the pose. See ref. 15 for additional details.