Virtual Exploration of Chemical Space by Database Generation

Virtual Exploration of Chemical Space by Database
Generation
Jean-Louis Reymond
University of Berne
Switzerland
1

The Chemical Universe Database (GDB)
> Chemical Space Travel
Swiss National Science Foundation
Tobias Fink
2

The Chemical Universe Project
HN
O
NH
S
J. Lederberg, C. Djerassi et al. J. Am. Chem. Soc. 1969, 91, 2973
T. Fink et al. Angew. Chem. Int. Ed. 2005, 44, 1504-1508, , J. Chem. Inf. Model. 2007, 47, 342-353
3

Structure Generator
Graphs from GENG
I. Graph Selection
II: Structure Generation
III: Filters
IV: Stereoisomers
843,335
↓
15,726
↓
276,220
↓
1.7 billion
↓
26.4 million
GDB
4

I. Graph Selection
Table 1. Graph selection table. Each graph corresponds to a saturated hydrocarbon.
Nodes Graphs a Passed Topo I b Passed Topo II c Planar Graphs d Unstrained Graphs e with unsaturations
1 1 1 1 1 1 1
2 1 1 1 1 1 3
3 2 2 2 2 2 4
4 6 4 4 4 4 13
5 21 8 8 8 8 33
6 78 20 20 20 20 123
7 353 57 57 57 57 445
8 1’929 199 194 194 194 1'956
9 12’207 780 712 708 705 8'863
10 89’402 3’600 2’893 2’845 2’822 43'443
11 739’335 19’215 12’575 12’169 11’912 221'336
Total 843’335 23’887 16’467 16’009 15’726 276'220
a generated by the program GENG using a maximum connectivity of 4. b The Topo I filters eliminates any graph with a
node present in two different 3 or 4-membered ring. c The Topo II filter eliminates graphs with a tetravalent bridgehead
in a small ring. d Non-planar graphs cannot be drawn in a plane without crossing edges, e.g. 3.
5

Tetrahedrane
I. Graph Selection
Claus Benzol
K 3,3 Graph
Tricyclo[2.2.2.2]decane
6

Count
350
300
250
200
150
100
50
0
1
6
Steric Energy Distribution over Ringsystems
11
16
21
26
I. Graph Selection
31
36
41
46
51
56
61
66
71
Highest Atomic Contribution [kcal/mol]
76
81
86
91
96
7

Graph automorphism
> Multiple bonds
– no allenes
– no DB at tetravalent nodes
– no DB at bridgeheads
– no TB at tri- or tetravalent nodes
– no TB inside rings smaller than 9
– no duplicates (symmetry)
> Elements
II. Structure Generation
– C, N, O, F at chemically meaningful nodes (valency rules)
1.7 billion "molecules"
276,220
8

III. Filters
> "Problematic" heteroatom constellations (X-X)
> Hydrolytically labile FG's
> Tautomers + aromaticity
OH
OH
OH
N
H
O
N
N
OH
NH 2
O
F
O O O
F (O, N)
O
OH
-CHO + -NH 2
9

Structure Generator
Graphs from GENG
I. Graph Selection
II: Structure Generation
III: Filters
IV: Stereoisomers
843,335
↓
15,726
↓
276,220
↓
1.7 billion
↓
26.4 million
GDB
10

IV. Stereoisomers
– Nourse, J. G.; Carhart, R. E.; Smith, D. H.; Djerassi, C. Exhaustive
Generation of Stereoisomers for Structure Elucidation. J. Am. Chem. Soc.
1979, 101, 1216-1223.
11

Table 2. Overview of the structure generation process.
Nodes Graphs a Generated b
GDB Overview
Accepted c
Unique
Tautomers
(GDB) d
All Tautomers Stereoisomers e
1 1 4 4 4 4 4
2 1 10 9 9 9 9
3 2 52 20 20 21 20
4 4 332 80 80 88 87
5 8 2’294 357 352 397 469
6 20 18’066 1’906 1’850 2’135 2’911
7 57 154’542 10’953 10’568 12’438 19’904
8 194 1’445’073 69’563 66’706 79’899 153’601
9 705 14’213’741 464’402 444’313 540’002 1’258’963
10 2’822 146’004’340 3’259’036 3’114’041 3’827’907 10’898’065
11 11’912 1’558’491’448 23’875’101 22’796’628 28’240’425 98’645’474
Total 15’726 1’720’329’902 27’681’431 26’434’571 32’703’325 110’979’507
log N = a x n 2
12

13
GDB by Graph
Monocycli
c
43%
Polycyclic
1%
Tricyclic
9%
Bicyclic
32%
Acyclic
15%
1
10
100
1'000
10'000
100'000
1
250
500
750
100
125
150
175
200
225
250
275
300
325
350
375
400
425
450
475
500
525
550
575
Graphs
Number of compounds
Acyclic
Monocyclic
Bicyclic
Tricyclic
Polycyclic
CH 4

Most Prolific Graphs
Bicyclic
32%
Tricyclic
9%
Polycyclic
1%
94'903 cpds 64'641cpds 48'836 cpds 12'983 cpds
Acyclic
15%
Monocycli
c
43%
14

GDB by Elemental Composition
Table 3. Number of compounds in GDB as a function of elemental composition (rows) and number of heavy atoms
(columns).
Number of heavy atoms
Elemental
composition
1 2 3 4 5 6 7 8 9 10 11 Total
C 1 3 4 12 29 102 347 1'468 6'413 30'582 152'117 191'078
N 1 0 0 0 0 0 0 0 0 0 0 1
O 1 1 0 0 0 0 0 0 0 0 0 2
F 1 1 0 0 0 0 0 0 0 0 0 2
CN 0 1 6 22 91 443 2'255 12'832 76'063 472'756 3'049'435 3'613'904
CO 0 2 5 19 74 338 1'671 9'302 54'733 337'024 2'164'860 2'568'028
CF 0 1 3 12 39 151 622 2'954 14'598 77'225 429'664 525'269
CNO 0 0 2 11 82 526 3'578 24'858 176'888 1'299'010 9'819'032 11'323'987
CNF 0 0 0 2 17 122 818 5'594 39'052 279'803 2'059'976 2'385'384
COF 0 0 0 2 18 126 809 5'437 37'148 260'489 1'883'487 2'187'516
CNOF 0 0 0 0 2 42 468 4'401 39'910 358'528 3'236'049 3'639'400
Total 4 9 20 80 352 1'850 10'568 66'846 444'805 3'115'417 22'794'620 26'434'571
15

Novelty vs. RDB
> Ring systems
> Stereochemistry
> Physico-chemical properties
> Compound classes
> Drugs and leads
IV: Stereoisomers
GDB Analysis
GDB
16

RDB = 63'857 cpds
Novelty vs. RDB
– organic molecules up to 11 atoms from PubChem, ChemACX, ChemSCX,
NCI open database, Merck Index
26'464
– Sulfur (14'048)
– Acyl halides etc. (7'782)
– S, P, Si (3'787)
– Allenes etc. (629)
– Topology (218)
RDB
37'393
GDB
26.4 million
17

Ring Systems
Number of 4-membered rings
Number of 3-membered rings 0 1 2 Total
0 124 [3] 189 [60] 103 [67] 416 [130]
1 225 [50] 238 [177] 20 [19] 483 [246]
2 201 [88] 55 [48] - 256 [136]
3 53 [26] - - 53 [26]
Total 603 [167] 482 [285] 123 [86] 1'208 [538]
18

Ring Sytems
– Chiral: 677 (367 unknown)
– Achiral: 455 (150 unknown)
– Both 76 (21 unknown)
> Acyclic graphs
– Chiral: 166 (all known)
– Achiral: 143 (all known)
> All graphs
Stereochemistry
– Chiral: 12'771
– Achiral: 2'955 19.2 million
Chiral
7.1 million
Achiral
134’390 (including tartaric acid!)
Fraction of SCU Compounds
100%
75%
50%
25%
0%
Chiral
Achiral
1 3 5 7 9 11 13 15 17 19
Number of heavy atoms
19

Stereochemistry
Table 5. Number of stereocenters vs. number of stereoisomers for GDB compounds.
Number of stereocenters a
Number of
stereoisomers b 0 1 2 3 4 5 6 7 8 9 10 Total
1 3’805’751 c
0 127’959 e
0 25’340 186 10’245 68 1’125 58 66 3’970’798
2 2’193’503 d 4’023’245 1’160’206 189’290 317’283 33’036 88’602 7’945 7’488 228 158 8’020’984
3 2’481 d
0 13’182 0 2’493 0 115 0 31 0 0 18’302
4 572’362 d 1’664’209 3’524’711 2’137’679 380’906 320’836 53’880 43’573 3’240 889 5 8’702’290
5 0 0 0 0 61 0 14 0 0 0 0 75
6 229 d
38 1’860 1’253 422 87 256 14 3 0 0 4’162
7 0 0 0 0 6 0 0 0 0 0 0 6
8 49’980 d
188’701 733’641 1’869’820 1’265’196 362’856 93’685 19’829 3’794 130 0 4’587’632
10 62 d
0 300 0 1’537 52 521 12 26 0 0 2’510
12 0 0 142 164 510 44 4 0 0 0 0 864
16 991 d
3’822 29’787 165’346 443’339 286’119 108’252 8’965 1’096 89 0 1’047’806
20 4 d
0 0 0 6 0 18 0 1 0 0 29
24 0 0 0 0 14 107 0 0 0 0 0 121
32 19 d
0 118 2’208 11’041 41’237 17’192 5’605 134 1 0 77’555
36 0 0 0 0 0 0 42 0 0 0 0 42
64 0 0 0 0 0 186 1’209 0 0 0 0 1’395
Total 6’625’382 5’880’015 5’591’906 4’365’760 2’448’154 1’044’746 374’035 86’011 16’938 1’395 229 26’434’571
a) A stereocenter is defined as an atom at which the interchange of two groups produces a stereoisomer, with the
exception of E/Z isomers. b) Total number of stereoisomers resulting from a compound. c) Molecules containing no
stereocenters and no E/Z double bonds. d) Molecules resulting in E/Z isomers only.
20

Stereochemistry
21

10 stereocenters
1 stereoisomer
Stereochemistry
6 stereocenters
64 stereoisomers
22

Novelty vs. RDB
> Ring systems
> Stereochemistry
GDB Analysis
> Physico-chemical properties
> Compound classes
> Drugs and leads
IV: Stereoisomers
GDB
23

Frequency x10 3
7000000
6000000
5000000
4000000
3000000
2000000
1000000
> MW ~ 153 Dal
> H-bond Acceptors ~ 3
> H-bond Donors ~ 1.5
> log P, NRB, TPSA
0
0
1
2
3
4
Physico-chemical properties
10000
5
8000
6000
4000
2000
6
Number of Rotatable Bonds
0
0
2
4
7
6
8
8
Frequency x10 3
Frequency x10 3
4500000
4000000
3500000
3000000
2500000
2000000
1500000
1000000
500000
0
2500000
2000000
1500000
1000000
500000
0
-12.0
0
-10.5
12
-9.0
24
-7.5
36
-6.0
48
-4.5
60
-3.0
72
-1.5
0.0
logP
84
96
1.5
4000
3000
2000
1000
108
6000
5000
4000
3000
2000
1000
0
-12
-9
-6
-3
0
3
6
9
3.0
120
4.5
132
6.0
144
Topological Polar Surface Area [A 3 ]
0
0
28
56
84
112
7.5
140
156
9.0
168
168
196
24

RO5 (Lipinski )
RO3 (Congreve)
Physico-chemical properties
MW
≤ 500
≤ 300
logP
≤ 5
≤ 3
Lipinski et al., Adv. Drug. Deliv. Rev. 1997, 23, 3
Congreve et al., Drug Discov. Today 2003, 8, 876
HBD
≤ 5
≤ 3
HBA
≤ 10
≤ 3
All structures OK
Half of all structures OK
NRB
≤ 3
TPSA
≤ 60 Å 2
25

Rdb
GDB
Property Space
DMU
26

Novelty vs. RDB
> Ring systems
> Stereochemistry
> Physico-chemical properties
> Compound classes
> Drugs and leads
IV: Stereoisomers
GDB Analysis
GDB
27

AC
d
N
p
d
where
=
=
=
⎧1
δ ij = ⎨
⎩0
N
= ∑∑ δ ⋅
i=
1 j=
1
Number of
if
if
N
Autocorrelation Descriptors
( p ⋅ p )
Considered topological
Atomic property
d
d
ij
ij
i
=
≠
atoms
d
d
j
d
distance
(Kronecker Delta)
Partial σ- and π-charges
Atomic polarizability
Topological steric effect index
Atomic number
Identity function
0-7 bonds
Moreau, G.; Broto, P. Autocorrelation of Molecular Structures. Application to SAR Studies.
Nouv. J. Chim. 1980, 4, 757-764.
28

48-dimensional AC vector
> 200x200 neurons, toroidal
> 1’000’000 GDB molecules
Kohonen Map
> 100 molecules/epoch, 250’000 epochs
> map 26.4 million
> Color code for properties
29

1 10 100 1000 10000
Occupancy
1 10 100 1000 1000030

Heavy atoms:
< 7 7 8 9 10 11
Molecular size
Mixed Empty
1 10 100 1000 1000031

Compound Classes
Heteroaromatic Aromatic Fused heterocyclic
Fused Carboc. Heterocyclic Carbocyclic
Heteroacyclic Carboacyclic Mixed Empty
1 10 100 1000 1000032

Lead-likeness and Chirality
Heteroaromatic Aromatic Fused heterocyclic
Carbocyclic Heterocyclic Carbocyclic
Heteroacyclic Carboacyclic Mixed Empty
Chiral
Achiral Empty
Lead-like
33
Mixed
Mixed
Not lead-like Empty

OH
O
O
O
Fragrances
Heteroaromatic Aromatic Fused heterocyclic
Fused Carboc Heterocyclic Carbocyclic
Heteroacyclic Carboacyclic Mixed Empty
HO
HO
O
O
O
O
O
34

O
O
O
O
O
O
O
O
O
O
Camphor Analogs
O
O
O
O
O
O
O
O
O O
O
O
O
O
O
O
O
O
O
O
O
35

O
O
O
O
O
O
Rose Oxide Analogs
O O
O
O
O O
O
O
O
O O
O
O
O O
O
O
O
O
O
O
O
O
O
O
O
36

Novelty vs. RDB
> Ring systems
> Stereochemistry
> Physico-chemical properties
> Compound classes
> Drugs and leads
IV: Stereoisomers
GDB Analysis
GDB
37

O
H 2N
N
H
39
(Paracetamol)
44
(GABA)
O
O
O
49
(Brevicomin)
OH O
OH
HO
H 2N
Known Drugs in GDB
OH
40
(Pregabalin)
O
NH 2
O
45
(Glutamate)
O
50
(Carvon)
OH
H 2N
NH 2
41
(Amantadine)
46
(Dopamine)
OH
51
(Menthol)
OH
OH
O
H
F
O
N
H
NH
O
42
(Fluorouracil)
HO
O
47
(Niacin)
52
(Benzaldehyde)
N
O
H
F
F
F
F
O F
F
43
(Desflurane)
O
O
48
(Frontalin)
53
(Cinnamaldehyde )
38

O
N
N
NH
N
N
N
H
F
F
N
O
N N
HN
N HN
H
N
OH
NH 2
Bayesian Virtual Screening
Heteroaromatic Aromatic Fused heterocyclic
Carbocyclic Heterocyclic Carbocyclic
Heteroacyclic Carboacyclic Mixed Empty
OH
N
HN
OH
37 38 39
O
OH
N
N
31 32 33
34 35 36
F
F
N
OH
N
NH
37
32
31
35
34
39
Kinase Inhibitors
GPCR ligands
Ion-channel modulators
36
33
38
39

The Chemical Universe Database (GDB)
> Chemical Space Travel
Swiss National Science Foundation
R. Van Deursen, J.-L. Reymond ChemMedChem 2007, 2, 636-640
Ruud van Deursen
40

Chemical Space Travel
> Move continuously via nearest neighbours
> Unknown space
> Structural vs. property space
A
B
n
41

A
Chemical Space as a Graph
B
Nearest neighbour mutations
Atom type exchange
Atom inversion
Atom removal
Atom addition
Bond saturation
Bond unsaturation
Bond rearrangement
42

Selection
Selector
Spaceship
Start A
Generator
Mutants
Ranking Target B
Ranked Mut.
No Target B found
Yes
Stop
Store in DB
43

A
A
Spaceship
B
mutation selection
mutation
B
F(c) = (T SF(target, c) x T PF(target, c)) 5
A
A
selection
B
B
44

A
Trajectory Libraries
B
Filters
"crude" possible
A
B
45

Compound Formula mass n [a]
Trajectory Examples (1)
Nearest
neighbours
Steps from CH 4 [b]
With
aromatic [c]
Steps to
N [d] MeOH [b]
Cubane C 8H 8 104 8 12 - 6'638 7 994
Fluorouracil C 4H 3FN 2O 2 130 9 16 9* 2'456 7* 560
Metheneamine C6H12N4 140 10 12 - 6'157 9 1'768
3-Tetrazene-2carboximidamide
C2H6N10 170 12 12
-
4'685 11* 2'007
Aspirine C9H8O4 180 13 15 8(1) 2'567 12 2'582
9-Ethyl-carbazole C 14H 13N 197 15 n.f. [e] 20(2) 20'501 16 5'357
Vitamin H C 10H 16N 2O 3S 244 16 18* - 27'161 14* 6'304
VX (van) C 11H 26NO 2PS 267 16 21 - 29'460 14* 3'954
Adenosine C 10H 13N 5O 4 267 19 n.f. [e] 25(2) 23'680 19* 13'639
β-estradiol C 18H 24O 2 272 20 23 15(2) 43'089 20 19'067
Retinal C 20H 28O 284 21 23 - 45'176 19* 15'100
Morphine C 17H 19NO 3 285 21 26 18(2) 69'113 20* 16'247
Aspartame C 14H 18N 2O 5 294 21 303 16(2) 34'172 20* 11'430
Cocaine C 17H 21NO 4 303 22 n.f. [e] 20(2)* 70'807 22 17'993
N [d]
46

Compound Formula mass n [a]
Trajectory Examples (2)
Nearest
neighbours
Steps from CH4 [b]
With
aromatic [c]
Steps to
N [d] MeOH [b]
Tetrodotoxine C 11H 17N 3O 8 319 22 28 - 106'158 20* 16'757
Sucrose C 12H 22O 11 342 23 25* - 67'052 21 19'552
Penicillin G C 16H 18N 2O 4S 334 23 n.f. [e] 20(2) 70'497 23* 15'748
Strychnine C 21H 22N 2O 2 332 25 n.f. [e] 26(2) 176'721 25 32'479
Papaverin C 20H 21NO 4 339 25 n.f. [e] 25(3) 53'099 25 28'449
Colchicine C 22H 25NO 6 399 29 37 32(3) 136'519 28 33'592
Calcitriol C 27H 44O 3 417 30 37* - 298'327 28* 65'595
Dipicrylamine C 12H 5N 7O 12 439 31 n.f. [e] 21(2) 21'015 26 13'950
Tetracycline C 22H 24N 2O 8 428 31 36 30(1) 173'734 30 34'883
Vitamin K C 31H 46O 2 451 33 55 42(3) 411'107 32* 77'337
Epothilone C 27H 41NO 6S 508 35 n.f. [e] 62(4) 709'250 34* 75'219
Vitamin E C 29H 50O 2 531 38 71 40(2) 443'477 37* 140'017
Reserpine C 33H 40N 2O 9 609 44 n.f. [e] 68(5) 286'342 62 230'646
Taxotere C 45H 55NO 15 808 58 n.f. [e] 74(4) 1'128'960 57* 304'172
N [d]
47

From: To:
Cubane
Aspirine
Cross-Trajectories
VX
Cubane - 10 18
Adenosine
23
(1)
Aspirine 10* - 14 21 15 16 24 22 22 33
VX 13
17
(1)
-
31
(1)
Adenosine 17* 27 18* - 14 15 24 23 27* 29
Sucrose 18*
22
(1)
22*
29
(1)
Sucrose
19
18
Penicillin G
18
(1)
15
(1)
- 25
Penicillin G 19* 13* 14* 23 19* - 20 19* 21* 29
Strychnine 21* 17* 20 26 22 16* - 30* 17* 22*
Colchicine 27 22* 21 26 18 22 23 - 22* 21*
Tetracycline 28* 20 25* 49 19 19* 16 28 - 17
Vitamin K 30* 24* 30* 34* 28* 27* 19* 30* 22* -
Strychnine
18
(1)
21
(1)
26
(1)
Colchicine
22
(1)
20
(2)
31
(1)
Tetracycline
24
(1)
24*
(1)
25
(1)
Vitamin K
26
(1)
25*
(1)
25
(1)
48

HO
N O
O
NH 2
OH
AMPA ↔ CNQX
14 ± 3
16 ± 2
O
H
O N N+
O-
O
N
H
N
49

N
O
N
OH
H 2N
AMPA
HO
HN
O
O
OH
NH
OH
O
O
NH 2
O O
HO N
O
NH
N
NH 2
O
O
N +
-
O
O
H
N
Trajectory
O
HO
OH
O
NH 2
O
1 2 3 4
OH
N
5 6 7 8
O
N
H
N O
OH
9 10 11
N
H
O
O
HN
O
O
O
O
OH
NH
NH
O
NH
N
NH 2
O -
N +
O
N
OH
HO
O
O
O
N
H
N
N
H
O
O
NH
N
H
CNQX
OH
N
H
N O
OH
O -
N +
O
N
50

AMPA-CNQX Libraries
> AMPA → CNQX
> CNQX → AMPA
> Runaway from AMPA
> Runaway from CNQX
(500 runs)
559,656 cpds
(20 steps, 100 runs)
152,916 cpds
51

Rearrangement
40%
Bond type operations
Desaturation: 0.735 ± 0.054
Saturation: 0.437 ± 0.274
Rearrangement: 0.512 ± 0.260
Atom type operations
Insertion: 0.788 ± 0.210
Removal: 0.550 ± 0.268
Identity: 1.000 ± 0.000
Inversion: 0.768 ± 0.062
Substitution: 0.726 ± 0.149
Mutation Statistics
Identity 1% 1% Inversion
Desaturation 1% 7% Saturation
24%
13%
13%
Insertion
Removal
Substitution
52

Docking
Table 4. Average estimated binding energies by docking of trajectory compounds using AutoDock.
Trajectory Ngenerated
[a]
NSelected SMILES
[b]
NStereoisomers BE (kcal/mol) [c]
AMPA to CNQX 353'036 3’571 8’570 -8.8 ± 0.9
CNQX to AMPA 206'622 2’051 5’754 -10.1 ± 1.0
run-away from AMPA 91'417 992 7’586 -9.1 ± 0.9
run-away from CNQX 61'499 974 5’751 -8.5 ± 0.9
[a] Randomly selected structures (as SMILES codes) from the trajectory library. [b] All possible stereoisomers were
generated using CORINA. [c] Average Binding Energy of the most stable bound conformation located by AutoDock.
53

HO
N O
O
NH 2
OH
Docking
H
N
N
H
1 (BE = -10.4) 2 (BE = -9.3)
HO
H 2N
O
O
OH
3 (BE = -13.9)
O
O
N
NH 2
O
N+
O-
N
54

1FTK.pdb
Docking
55

Docking, synthesis and testing
> Expand GDB
> Graphical user interface
Outlook
Swiss National Science Foundation
56