Transcript pinkstuff.publication.org.uk

Knowing When
to Give Up
Sensible Evaluation
in Virtual Screening
for Drug Discovery
Gwyn Skone
Stephen Cameron and Irina Voiculescu
Virtual Screening
• Part of the Drug Discovery ‘Pipeline’
• Identify & validate target
• Prepare ligand database
 Filter ligands to select leads
• Lead result optimization
• Pre-clinical testing
• Clinical trials
• Approval and marketing
Virtual Screening
How is it done?
Target molecule
Set of Ligand
Conformations
Target molecule
Ligand Pose
Dock
Set of Scored Poses
Score
Real
How is it done?
Dock(T, Ls) =
• For each ligand LLs...
• For each possible conformation...
• For each considered pose...
• Calculate Score(T, L)
Score(T, L) =

lLatoms

tTatoms
N

i=1
Fi(t, l)
How might we improve it?
• Function-Based Early Rejection
-150
31
29
27
25
Atoms evaluated
23
21
19
17
15
13
11
9
7
5
-325
3
1
Threshold
score
50%
45%
40%
35%
30%
25%
20%
15%
10%
5%
0%
Percentage of poses rejected at
each stage (atom) of scoring function
evaluation, by rejection threshold
-500
How might we improve it?
• Prioritized Atom Evaluation
-150
31
29
27
25
Atoms evaluated
23
21
19
17
15
13
11
9
7
5
-325
3
1
Threshold
score
50%
45%
40%
35%
30%
25%
20%
15%
10%
5%
0%
Percentage of poses rejected at
each stage (atom) of scoring function
evaluation, by rejection threshold
with the use of atom ordering
-500
How might we improve it?
• Red: percentage of poses abandoned;
• Blue: reduction in run time.
• Dotted lines are the unordered case.
Early-Rejection Effect Statistics
60%
40%
20%
0%
None
-25
-50
-75
-100
-125
-150
-175
-200
-225
-250
-275
-300
-325
-350
-375
-400
-425
-450
-475
-500
Abandoned Poses /
Time Improvement
80%
Threshold Score
How might we improve it?
• Result Quota Consideration
• Black line: run time; Bars: results by RMSD (Å):
0  Blue  1 < Green  2 < Yellow  3 < Red  4Å
Quota-Based Rejection Thresholds
(averaging 3 runs of 200 results)
50%
5
30%
3
20%
2
10%
1
0%
0
Minimum number of generations
90
84
78
72
66
60
54
48
42
36
30
24
18
12
6
All
Results
4
Mean run time
(minutes)
40%
Is there anything more positive?
• Use Quaternions to represent rotations
50%
12
40%
10
30%
8
20%
6
10%
4
0%
2
Euler Angles
Quaternions
Run time (minutes)
Results
Comparison of Rotation Representations
Is there anything more positive?
• Learn From Prior Experience
• Search Pose Pre-Selection
Is there anything more positive?
• Start in Likely Places
• Automatic Pocket Detection
end.