SAESAR Shape And Electrostatics in SAR

1
SAESARShape And Electrostatics in SAR
Norah E. MacCuish, Anthony Nicholls, John D.
MacCuish
CUP V Santa Fe, NM, Monday, March 1, 2004
OpenEye Scientific Software, Inc
2
Introduction

• Shape and Electrostatics What role do they play
  in determining Structure Activity Relationships ?
• Is there a protocol using OpenEye and Mesa
  software that can be applied to drug design
  problems which takes advantage of Shape and
  Electrostatics?

3
Outline

• Shape Electrostatics - generation and
  comparisons
• Types of problems which will be discussed
• Bound ligand data- Xray (Protein Data Bank)
  versus assay ligand data (Wombat Database
  Sunset Molecular Discovery, LLC ) and decoys
• Cox2
• Progesterone
• Assay Ligand Data (Wombat Database) versus decoys
• Dopamine
• Ca Ion Channel
• Wombat Database contains ligands with published
  activities across many receptor types.

4
Tanimoto Measure For Shape and Electrostatics

• Tanimoto Shape Comparison for a,b volume
  overlaps
• Tab Oa,b/(Ia Ib- Oa,b) range (0,1)
• Tversky Shape Comparison for i,j (subshape)
• Ta,b Oab/(aIa bIb - Oab) range(0,1)
• Tanimoto Electrostatic Comparison for a,b field
  overlaps

range(-1/3,1) (Using MMFF charges, continuum
solvent)
OpenEye Scientific Software
5
Shape Tanimoto Example
Tanimoto 0.874
ROCS, Shape Toolkit
OpenEye Scientific Software, Inc
6
Does a Maximum Shape Match Assure a Maximum
Electrostatic Match? NO! (EON Spin)
xray structure
Shape Tanimoto 0.956 Electrostatic Tanimoto0.108
Shape Tanimoto0.942 Electrostatic Tanimoto0.293
Conformers of an active ligand - Omega
Electrostatic Overlay - Eon
Eon Omega
7
Electrostatic Value Conformer Choice Based on
Geometric Mean (of Electrostatic Tanimoto and
Shape Tanimoto) or Maximum Shape
Geometric Mean Maximum Shape 40 of the time
60 of the time, using the Geometric Mean the
Electrostatic Tanimoto larger with only a small
change in Shape Tanimoto
8
Conformational Energy Analysis using Xray
structure and Wombat HIVRT ligands

• All conformer similarity at 3kcal, 5kcal, 10kcal
• Geometric Mean conformer similarity at 3kcal,
  5kcal, 10kcal
• For an xray structure of a HIVRT ligand versus
  ligands active for HIVRT from Wombat.

Xray structure
9
HIV-RT Actives All conformer similarities at 3,
5, 10 kcal
10
HIV-RT Actives Geometric Mean conformer
similarity at 3,5 and10 kcal
11
Electrostatic Tanimoto vs RMS from Crystal
Structure After aligning to MKC-442 (HEPT
analogues)
1rt2
Poor conformation (Shape Tani lt 0.6)
Conformation match (Shape Tani gt 0.88)
1rti
1c1b
MKC-442
1c1c
1rti
1c1b
1c1c
1rt2
12
Crystal Structure of Target with bound ligand vs
reported actives and decoys

• Generate conformations of Wombat ligands at 5kcal
  with Omega
• Transform conformer space and electrostatic space
  to single conformer and electrostatics per
  structure vs. xray structure (Using Geometric
  Mean)
• Combine with 2D structure space
• Develop simple model against decoys, validate
• Variables (shape, electrostatics, 2D (320 MACCS
  keys))

13
43 Cox2 Ligands - Highly Active Structures from
Wombat Shape and Electrostatic Tanimoto
Similarities to Cox Crystal Ligand
Xray structure Cox2 SC-558
14
Decoys plus Highly Actives Shape and
Electrostatic Tanimoto Similarities to Cox
Crystal Ligand
Xray structure Cox2 SC-558
15
Decoys plus Highly, Moderately, and Weakly
Actives Shape and Electrostatic Tanimoto
Similarities to Cox Crystal Ligand
Xray structure Cox2 SC-558
16
Cox2 Classification Error

class means
0.81 STD of error 0.03 Classification error
Geo-mean2 For uneven Class sizes
100 fold Cross validation Testing set Randomly
sampled Data Divided into 3/5 Training, 2/5
Testing set
Linear Decision Boundary (Fishers Linear
Discriminant)
17
Enrichments and Classification Error
18
Progesterone Receptor with bound Progesterone
FASP
19
Progesterone Study
20
Progesterone Receptor
FASP
21
Group Average Clustering of All (100 SMILES)
Progesterone Actives Cut made at 0.68 similarity
to pick up 25 of SMILES in Cluster and 65 of
SMILES in Cluster B
Grouping Module Shape Module
12 remaining outliers
0.68 Similarity
Cluster A (24 SMILES)
Cluster B (64 SMILES)
Progesterone
Cluster A centroid
Cluster B centroid
22
Cluster A (Conformers chosen via the geometric
mean of Shape and ET values)
Cluster B (Conformers chosen via the geometric
mean of Shape and ET values)
23
Cluster A and B Centroids vs Wombat Decoys and
Cluster Members
Cluster A Centroid
Cluster B Centroid
24
Cluster A, comparing 2D Tani with ET and Shape
Cluster B, comparing 2D Tani with ET and Shape
25
Decoys plus Highly, Moderately, Weakly
Actives Shape and Electrostatic Tanimoto
Similarities to Cox Crystal Ligand
Xray structure Cox2 SC-558
(When actives in outlier region are removed the
Geo-mean error 0.95)
26
Cox Outlier Study

• Outliers
• Moderately High Shape Similarity to COX2 SC-558
• No Electrostatic Similarity to SC-558
• Found Strong Shape (gt 0.75) Cluster that Covers
  all Structures
• Similarity to Centroid (or representative
  compound) of Cluster Shows
• Most have moderate to strong ET similarity
• Wide Spread of 2D Structure.

27
Xray Structure compared with outlier cluster
centroid
COX2 SC-558
Centroid
28
Outlier Cox Ligands versus Outlier Centroid
29
Cox2 Outlier Ligands vs. Outlier Centroid 2D
structure analysis
30
Summary of xray data analysis

• Shape and Electrostatics from xray queries will
  classify actives versus decoys
• Shape clustering provides a way of exploring
  anomalies, other binding modes, etc. unexplained
  by the classification model

OpenEye Scientific Software, Inc
31
Dopamine D2 Ligands

• Clustering of 28 active structures (4188
  conformers) with Taylor-Butina clustering at 0.7
  Tanimoto threshold.
• First three largest clusters contain 20 (455
  conformers), 18 (210 conformers), and 18 (185
  conformers) structures respectively, but the
  union of these three clusters contained 22 of the
  28 structures, thus much overlapping.
• 1000 Wombat Decoys were used against each
  centroid for each of the three clusters.

32
Dopamine D2 Cluster 1 Centroid vs Highly Actives
and Wombat Decoys
3 active structures
Centroid
33
Dopamine D2 Cluster 2 Centroid vs Highly Actives
and Wombat Decoys
2 active structures
Centroid
34
Dopamine D2 Cluster 3 Centroid vs Highly Actives
and Wombat Decoys
1 active structure
Centroid
35
Wards Clustering of 14 Ca Ion Structures with
185 Conformers
All but one structure represented
Remaining conformers from just one structure,
not represented in group on left
Cluster 1 One centroid 6 other compounds
Cluster 3 One centroid 3 other compounds
Cluster 2 Just 1 compound
Cluster 4 Just 1 compound
36
Example Shape Comparison with Tversky
Tversky .974
Tversky .974
Tversky .872
Tversky .872
How well does A fit into B and How well does B
fit into A
37
Ca Ion Channel Overlay Result
Representative Shape
One finding - An Inflexible conformer which is
contained in at least one conformer for every
SMILES in the input dataset
Shape Overlay for Flexible Ligand Conformers
with Tversky gt.85
Investigations in shape clustering for lead
hopping, NE MacCuish, JD MacCuish, 226 ACS New
York Sept. 10, 2003
38
Ca Ion Channel Subshape and Electrostatic
Overlays
ETani.18 STversky0.99 Shape Tani0.78 2D
Tani0.58
ETani.15 STversky0.91 Shape Tani0.71 2D
Tani0.40
ETani.18 STversky0.99 Shape Tani0.76 2D
Tani0.62
ETani.55 STversky0.96 Shape Tani0.85 2D Tani
0.89
39
Conclusions

• Shape and Electrostatics are discriminating
  descriptors for predictive modeling of activity
  relationships
• A protocol which combines OpenEye and Mesa
  software can facilitate SAR analysis
• Starting with SMILES, 3D shape and electrostatic
  patterns can be derived
• Shape, partial shape, and electrostatics are
  effective tools
• Negative information is significant.
• Future work more receptors, clustering on both
  shape and electrostatics

OpenEye Scientific Software, Inc
40
Acknowlegments

• OpenEye Scientific Software
• Roger Sayles
• Geoff Skilman
• Bob Tolbert
• Stanislaw Wlodek
• Jeremy Yang
• Tudor Oprea Sunset Molecular Discovery, LLC
• Protein Data Bank