Characterization of Pharmacophore Multiplet Fingerprints as

Characterization of Pharmacophore Multiplet Fingerprints as Molecular Descriptors Robert D. Clark Tripos, Inc.

[email protected]

©2004 Tripos, Inc.

Outline • Background o o

history mechanics

• Finding appropriate binning ranges o

biased conformer generation

• Similarity measures o

stochastic similarity

• Hypothesis generation o

asymmetric similarity

• Conclusions

History of Pharmacophore Multiplets A.C. Good and I.D. Kuntz; J. Comput.-Aided Mol. Design 1995, 9, 373-379. X. Chen, A. Rusinko, and S.S. Young; J. Chem. Inf. Comput. Sci. 1998, 38, 1054-1062. J.S. Mason, I. Morize, P.R. Menard, D.L. Cheney, C. Hulme & R.F. Labaudiniere; J. Med. Chem. 1999, 42, 3251-3264. M.J. McGregor & S.M. Muskal; J. Chem. Inf. Comput. Sci. 1999, 39, 569-574. H. Matter and T. Pötter; J. Chem. Inf. Comput. Sci. 1999, 39, 1211-1225. J.S. Mason and B.R. Beno; J. Mol. Graphics Mod. 2000, 18, 438-451 E. Abrahamian, P.C. Fox, L. Nærum, I.T. Christensen, H. Thøgersen & R.D. Clark; J. Chem. Inf. Comput. Sci. 2003, 43, 458-468.

Novo Nordisk / Tripos Tuplets Collaboration • 2 year collaboration to develop and extend existing SYBYL triplet (PDT) technology

• Incorporate pair, triplet and quartet (‘Tuplet) technology • Augmented ‘Tuplets and support for privileged substructures

• Conformers generated on-the-fly or retrieved • Bitmaps created, stored and manipulated in compressed format o

four 1.8 x 109 bit bitmaps stored as ~80kb file

o

0.01-0.5 seconds/molecule

Type III antiarrhythmic: UK 66914

donor atom

positive nitrogen

acceptor atoms

hydrophobic center

hydrophobic center donor/acceptor atoms

Multiplet Fingerprints

… 000010001010000000100100001110100001110000111000000000011001...

Indexing Triplets

2 Vertex joining longest and shortest edges

D

3

A

H 5

Bin: 5, 3, 2 Triplet: H-A-D

Indexing Tetrahedra Problems: • Need a unique mapping • Must deal with chirality • Literally dozens of possible permutations • Mapping must be based on bins and features C

2

2

D

D

4

Plane of symmetry implies no chirality

3 2 A

C

2

4 3 A

4

2

4 3

C

C

4

C

2 D 2 B

Chiral tetrahedra

A

4 C

D 2 B

4 3 4

A

Mapping Quartet Bits Mapping for 7 bins and 3 features (D, A, H)

000000 000001

...

542333*

DDDD DDDA DDDH

...

...

666665 666666

HHHH

Bitmap Size = 76 * 34 = 9,529,569 bits *542333 specifies the + enantiomer; 245333 specifies the - enantiomer

-

+

frequency

beta blockers

frequency

0

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

150

K+ channel openers

100 50 0 0

frequency

Distribution of Distances Between Features

300 250 200 150 100 50 0

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

200 150

Type I antiarrythmics

100 50 0 0

1

2

3

4

5

6

7

8

9

edge length (Å)

10

11

12

13

14

15

frequency

1800 1600 1400 1200 1000 800 600 400 200 0

Estrogen Antagonists Type III Antiarrythmics Benzamides Phenothiazines Beta Blockers Type I Antiarrythmics K Channel Openers 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

100 Conformer By Class 50000

frequency

Cumulative Distributions across Classes

1 Conformer By Class

Estrogen Antagonists

40000

Type III Antiarrythmics

30000

Benzamides Phenothiazines

20000

Beta Blockers

10000

Type I Antiarrythmics K Channel Openers

0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

edge length (Å)

frequency

50000

Estrogen Antagonists

40000

Type III Antiarrythmics

30000

Benzamides Phenothiazines

20000

Beta Blockers

10000

Type I Antiarrythmics K Channel Openers

0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

100 Systematic Search Conformers By Class

frequency

Effect of Biased Conformer Generation

100 Confort Conformer By Class

16000 14000 12000 10000 8000 6000 4000 2000 0

Estrogen Antagonists Type III Antiarrythmics Benzamides Phenothiazines Beta Blockers Type I Antiarrythmics K Channel Openers 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

edge length (Å)

Hypothesis Fingerprint Creation DDD DDD DDA DAA DDH DAH DHH HHH 000 001 200 210 210 331 333 433

Binary Compound Fingerprints

0

1

0

0

0

1

1

0

0

1

0

1

0

0

1

0

1

0

0

1

0

1

1

1

0

1

1

1

0

0

1

0



Vector Sum Fingerprint

0

1

0

0

0

1

1

0

0

1

0

1

0

0

1

0

1

0

0

1

0

1

1

1

0

1

1

1

0

0

1

0

1

3

1

3

0

2

4

1


0

1

0

0

0

1

1

0

0

1

0

1

0

0

1

0

1

0

0

1

0

1

1

1

0

1

1

1

0

0

1

0


1

3

1

3

0

2

4

1

Feature Weights

3

3

3

3

4

4

5

6

Bin Weights

3

4

5

6

6

10

12

13

Bit Score

9

36

15

54

24

80

240

78


Weighting Bits for Hypothesis Generation nf

nd

i =1

j =1

Sb = f b × ∑ fwi × ∑ dw j Sb is the score for the bit fb is the frequency of the bit fwi is the weight of the feature type dwj is the weight of the distance bin

⇒Construct an hypothesis from the highest scoring bits.

f1 d2

d1 f2

d3

f3


0

1

0

0

0

1

1

0

0

1

0

1

0

0

1

0

1

0

0

1

0

1

1

1

0

1

1

1

0

0

1

0


1

3

1

3

0

2

4

1

Feature Weights

3

3

3

3

4

4

5

6

Bin Weights

3

4

5

6

6

10

12

13

Bit Score

9

36

15

54

24

80

240

78


Sanity Checker

N tn S= Nt

Similarity Measures • Tanimoto coefficient t ( A, B) =

pdt ( A) ∩ pdt ( B) pdt ( A) ∪ pdt ( B)

• Cosine coefficient Cc (a, b) =

pdt (a ) ∩ pdt (b) pdt (a ) × pdt (b)

• Stochastic cosine coefficient s ( A, B) =

[

[

E pdt ( A) ∩ pdt * ( B )

] [

]

E pdt ( A) ∩ pdt * ( A) × E pdt ( B ) ∩ pdt * ( B)

]

Effect of Conformer Count on Stochastic Cosine Similarity

similarity

0.6 0.5

Estrogen_Antagonist Class Similarity

0.4

Estrogen_Antagonist Non-Class Similarity K_openers Class Similarity

0.3

K_openers Non-Class Similarity

0.2

benzamides Class Similarity

0.1

benzamides Non-Class Similarity

0 0

100 200 300 400 500 600 700 800 900 1000

conformer count (max)

Effect of Conformer Count on Stochastic Cosine Discrimination discrimination ratio

14.0000 12.0000

I_Antiarrythmics

10.0000

III_Antiarrythmics Phenothiazines

8.0000

beta Blocker

6.0000

Benzamides

4.0000

K_openers

2.0000

Estrogen_Antagonist

0.0000 1

10

100


1000

discrimination ratio discrimination ratio

Discrimination and Similarity Measure

14.0000

simple cosine

12.0000

I_Antiarrythmics III_Antiarrythmics

10.0000

Phenothiazines

8.0000

beta Blocker

6.0000

Benzamides

4.0000

K_openers

2.0000

Estrogen_Antagonist

0.0000 1

10

100

1000

20.0000

Tanimoto


15.0000

Phenothiazines

10.0000

beta Blocker Benzamides

5.0000

K_openers Estrogen_Antagonist

0.0000 1

10

100

1000


discrimination ratio

CONFORT

12.0000


10.0000

Phenothiazines

8.0000

beta Blocker

6.0000

Benzamides

4.0000

K_openers

2.0000

Estrogen_Antagonist

0.0000 1

discrimination ratio

Discrimiantion and Conformer Bias

14.0000

10

100

1000

14.0000

systematic search

12.0000


10.0000

Phenothiazines

8.0000

beta Blocker

6.0000

Benzamides

4.0000

K_openers

2.0000

Estrogen_Antagonist

0.0000 1

10

100


1000

Symmetric Similarity Measures • Symmetric stochastic cosine s ( A, B) =

[

[

E pdt ( A) ∩ pdt * ( B)

] [

E pdt ( A) ∩ pdt * ( A) × E pdt ( B) ∩ pdt * ( B )

• Asymmetric stochastic cosine s *(h, t ) =

]

[

E pdt (h) ∩ pdt (t )

[

]

E pdt (h) ∩ pdt *(h)

]

]

average similarity

0.6 0.5 0.4 0.3 0.2 0.1 0

CONFORT within class

symmetric cosine

100 Conformers without class 0

average similarity

Effect of Hypoothesis Size (Type III antiarrhythmics)

asymmetric stochastic cosine

200

0.6 0.5 0.4 0.3 0.2 0.1 0

400

600

800

1000

systematic search within class

1000 Conformers without class 0

200

400

600

bits in hypothesis

800

1000

Conclusions • Compression is cool • Natural binning does make sense o o

1.75 3 4 5 6 7 8 8.75 9.75 10.75 11.75 13 15 >15Å at least for triplets

• Systematic bias increases discrimination o o

rule-based conformational bias can be useful caveat: it may limit lead-hopping

• More is not necessarily better o o

true in terms of conformation count true in terms of multiplet hypothesis size

• A little asymmetry can be a good thing • Compression is still cool

Acknowledgements

www.tripos.com

Novo Nordisk A/S (Denmark) Lars Nærum* Henning Thøgersen* Tripos, Inc. Edmond Abrahamian Peter Fox Trevor Heritage

May the multiplets be with you...

What a Protein “Sees”

(electrostatic field at 0.5 Å resolution, 80 and 30% contours)

What the Chemist Sees H3C

O S

Cl

O

O

N

O

H3C

N N

O

F O

H3C

N

H3C

N H

CF3

O tetrahydrophthalimide (American Cyanamide)

trifluorotoluidide pyrazole ether (Monsanto)

Pharmacophoric Features hydrogen bond acceptors

H3C

O S

Cl

O O

N

O

H3C

N N

O

F

H3C

O

hydrophobic centers

N

O

H3C

hydrogen bond donor

N H

CF3

Conformational Sampling*

*diverse conformers obtained using CONFORT

Mapping Multiplets Mapping for 7 bins and 3 features (D, A, H)*

000

001

...

532

1 bit DDD

DDA

DDH

...

665

...

Bitmap Size = 73 * 33 = 9261 bits * Features are handled in the order supplied by the application.

666

HHH

Hypothesis Generation Multiple methods implemented for hypothesis generation o

From a collection of known actives

o

From a user defined UNITY® query

o

From a single molecule pharmacophore map a)

o

Single or multiple generated conformers

From user specified residues in receptor cavity

Privileged Substructures: Augmented Triplets DS HY

AA

HY @_AUGMENTED # name DONOR_SITE

. =NULL

.

mnemonic DS

xref AA

weight 3.0

min_dist 2.5

max_dist 3.5

Effect of Conformer Count on Cosine Coefficient Similarity

similarity discrimination ratio

0.6 0.5

Estrogen_Antagonist Class Similarity

14.0000 0.4

Estrogen_Antagonist Non-Class Similarity

12.0000

K_openers Class Similarity I_Antiarrythmics

0.3 10.0000

III_Antiarrythmics K_openers Non-Class Similarity Phenothiazines

8.0000

beta Blocker benzamides Class Similarity Benzamides

0.2

6.0000 4.0000

K_openers benzamides Non-Class Similarity Estrogen_Antagonist

0.1

2.0000 0.0000 0 0

1 200 300 10 100700 800 1000 100 400 500 600 900 1000
