A new knowledge-based approach to the de novo design of synthetically feasible molecules is described. The method is based on specifically designed transform libraries abstracted from reaction databases. The structure generation process is based on conceptual chemistry and the degree of complexity introduced in the new structures can be modulated using specific parameters. Furthermore, this new system allows the integration of the results obtained in different workflows to calculate/predict other important physico-chemical properties of the new suggested molecules.

1. Knowledge-based
Compound Design
Fernando F. Huerta
Chemnotia AB powered by
InfoChem GmbH

2. QSAR models
Ligand Based
Pharmacophore
models
Structure Based
HTS
Compound (Drug) Design
Fragment Based

3. How Drugs Look?
HN
N
N
N
O
S
O
OH
NH2 O
O
F
F
N
HN
N
O
OH
HN
N
NH2 O
F
F
N
O
HN
O O
N
O
OH
NH2 O
F
F
N
HN
HN
N
O
OH
O
F
F
N
HN
Norfloxacin
(Noroxin™)
Ciprofloxacin
(Cipro™)
Sparfloxacin
(Zagam™)
Grepafloxacin
(Raxar™)
N
Drug Discovery Today 2011, 16, 722
Drug Discovery Today 2011, 16, 779
N
S
O
N
O
O
N
S
O
N
F O
F
Nexium®
Rabeprazole®
PANTOPRAZOLE®
O
N
N
Cl S
N
N O
HO
Loxapine®
Seroquel®

4. Why Do Drugs Look Similar?
Protein
Binding
Site
Drug Design Paradigm
Activity, Selectivity,
ADMET,..
Similar structural motif
share similar properties(?)

5. Compound (Drug) Design
What would you do?

6. Knowledge-based
Compound Design

7. Transforms from Reaction Databases
Reaction database
e.g. SPRESI
proprietary databases
commercial databases
1. Pre-processing
Template (2 levels):
Transform:
Transform library
Automatic transform
extraction
(ICMAP/CLASSIFY)
N
Remove bond 1-3
1
11
R1
12
R2
10 9
2
6
3
4
O
5
7 8
13
14
Remove bond between atoms 3 and 6.
Make new single bond bReetwmeoevne abtoonmds 1 3- 2and 5
Decrease bond order of double bond 2=4 by 1.
Make new single bond between atoms 3 and 4
Make new single bond between atoms 2 and 6.
Add group to atom 3: -OH
2. Retrosynthesis
Lookup
stored
example
reactions
H
O
O
H
Target molecule
Precursor(s) 1
Precursor(s) 2
...
Precursor(s) n
Transform x
Transform y
...
Transform z

8. Transforms from Reaction Databases
Reaction database
e.g. SPRESI
proprietary databases
commercial databases
Transform library
2. Forward
Reaction
Lookup
stored
example
reactions
Starting Material
or Reactant
Transform x
Transform y
Transform z
Product(s) 1
Product(s) 2
...
Product(s) n
...
N
Cl N
O

9. Retrosynthetic
Analysis
• Target orientated
• Complexity reduction
• Availability of starting
materials
• Multistep process
Reaction
Prediction
• Unknown product
molecules
• Molecular size
• Reagents
• Single reaction step

10. Forward Reaction Prediction (cont.)
• Number of transforms
NH N 2
O
S
OH
O O
N
OH
H H

11. ICSYNTH Strategy Parameters
• Strategy: defined by a set of parameters
• Rating of generated precursors/products
• 42 parameters (Retrosynthesis / FRP)
ICSYNTH FRP Parameters
O OH
Reactant
Transform
Product(s)

12. Strategy Parameters Optimization
42 parameters
(reactant, transform, product)
optimization algorithm

13. ICSYNTH FRP Strategies
Reactivity Mapping
FG Library Design
Ring Introduction
Scaffold Modification: Not Scaffold Hopping!!

14. FRP Examples / Applications
Reactivity Mapping
N
N O
N NH
O
N
N O
N NH
O
R
N
O R
N N
N NH
N
N O
N NH
O
Hal
N
N O
N N
O
N
N O
O N NH
O
R

15. FRP Examples / Applications
FG Library Design
N
HO O
N NH
reaction center

16. Knowledge-based
Compound Design
Other Databases

17. Knowledge-based Drug Design
• One synthetic step (reliable?)
• Novelty
• Drug like
• Physicochemical properties
• Lipinski

18. Knowledge-based Drug Design
inspired by “de novo design using reaction vectors”
Example 1
HN
O N
S
OH
O
O
Penicillin G
Example
from
J.
Chem.
Inf.
Model.,
Vol.
49,
No.
5,
2009,
1163-­‐1184

19. Knowledge-based Drug Design
inspired by “de novo design using reaction vectors”
Example 1
HN
O N
S
OH
O
O
Penicillin G
Example
from
J.
Chem.
Inf.
Model.,
Vol.
49,
No.
5,
2009,
1163-­‐1184
HN
O N
S
OH
O
O
Reaction Vectors
ICSYNTH FRP (precision Medium)
ICSYNTH FRP (precision High)

20. Knowledge-based Drug Design
Example 1

21. Knowledge-based Drug Design
Example 1
• 30 suggested cpds from ICSYNTH vs 1000 cpds (Reaxys,
penicillinG sss)
• Fingerprint similarities calculated (30000)
• Identical compounds filtered off (similarity = 1)
• Compounds with Tanimoto distance between 0.5-0.9
were selected
• Synthetic data (ICSYNTH) available
• Calculated med chem properties of new suggestions as
well as med chem properties of previously reported
ones available for analysis

22. Knowledge-based Drug Design
Example 2
N
Cl N
O
Core Structure
(Diazepam)
Reaxys search
214 reactions / products
N
Cl
N
F
O
N
Cl
HO
O
N
O
N O
O
N
O
N
Cl
O
NH
O
NH
… Cl N
N O
100 suggestions

23. Knowledge-based Drug Design
Example 2
Processing Information
• 80 new reactions identified
• 10 Identical matches
• 8 suggested products with a Tanimoto
distance <0.2
• 2 abstraction errors (knime)
but… it’s not enough for compound design…

24. Knowledge-based Drug Design
Example 2 (part II)
N
Cl N
O
Core Structure
(Diazepam)
80 new suggested
reactions/products
versus
Reaxys substructure search
2564 related compounds

25. Knowledge-based Drug Design
Example 2 (part II)
• Fingerprints calculated
• Identical compounds filtered off (Tanimoto = 1)
• 13 suggested products with a Tanimoto 0.7-0.9
• Physicochemical properties calculated and compared
New “suggested” molecules show
similar properties to the known-ones

26. Knowledge-based Drug Design
Example 3
Cl
N
O
N
N
Cl
HN
O
O
Intermediate* used for
ICSynth FRP and Reaxys sss
Loxapine®
• 61 suggestions ICSYNTH FRP (one synthetic step away from intermediate)
• 922 related compounds in the literature (Loxapine included)
• 6 new compounds with Tanimoto distance between 0.5-0.9 were suggested
N
O
R
N
N
Cl O
* Not real precursor for the final molecule
N
N
Cl
Ar
Z
O
NH
Cl
Ar
O
X
Cl

27. One more thing…
building synthetic confidence
filtering ICSYNTH cpds with low synthetic background
• number of precedent reactions of the same type
• precedent yield reported
• level of similarity from the original reaction

28. Summary

30. ACKNOWLEDGMENTS
InfoChem
Peter Loew
Heinz Saller
Christoph Oppawsky
Mike Hutchings
Hans Kraut
Valentina Eigner-Pitto
Josef Eiblmaier
Ulf Frieske
Stephanie North
Chemnotia
Anders Bogevig
Tobias Rein
THANK YOU