Chemistry Reference
In-Depth Information
determine how many 'unique' poses exist, we have initially resorted to clustering the poses
according to their pairwise RMSDs and then evaluated the binding poses. Although the
nature of the systematic sampling done by
Poser
does not lend itself to standard clustering
algorithms, we have done so anyway on the premise that the pose distribution will not
be uniform once the pose set is filtered by the bump check, contact criteria and NOE
matching score.
As an example, we show the results for clustering the poses of the Bcl-x
L
/
7
test case.
There were 9556 poses generated by
Poser
that were evaluated by NOE matching. The
COST values from NOE matching ranged from 1383 to 5541, as shown in Figure 5.12D.
The target pose had a COST of 2015. Poses were clustered by RMSD using an intern-
ally developed average-linkage method. The intra-group average pairwise similarity was
required to be below a threshold of 0.8 Å. For these poses, the result was 1077 clusters,
of which 221 were singletons. This was still too many to visualize. By eliminating any
pose with a COST greater than two times that of the lowest COST pose, the number of
poses to be considered was reduced from 9556 to 156. These poses were clustered as
before, this time resulting in 44 clusters, of which 21 were singletons. The largest and
second-largest clusters contained 22 and 12 members, respectively, and differed signific-
antly in structure. A plot of the 10 most populated clusters versus their NOE matching
COST (Figure 5.13A) indicates that two of the clusters (clusters 2 and 7) contain mem-
bers that have significantly lower COST than all of the other clusters. The RMSD of
the poses representing the centroids of these two clusters indicates that the members of
the clusters are structurally similar; the RMSD is 1.14 Å. Moreover, if one plots the
(a)
3200
2800
2400
2000
1600
1200
0
40
80
POSE #
120
160
Figure 5.13
(A) The lowest COST poses for the NOE matching run on Bcl-x
L
/
7
were clustered
based on RMSD to each other. The individual poses for each of the 10 lowest COST clusters
were ordered based on COST and plotted from lowest COST to highest COST. Members of the
different clusters are indicated with distinct symbols. All of the lowest COST poses belonged
to one of two clusters, as shown in (B). (B) COST versus the RMSD (Å) to the cluster 2 centroid
for Bcl-x
L
/
7
. Member poses of the 10 lowest scoring clusters are indicated as in (A).
