1. A knowledge-based approach for reaction generation Development, validation and applications Dimitar Hristozov, 04.06.2009

2. Motivation public reaction databases >1,500,000 reactions covering general organic chemistry medicinal chemistslab notebooks (eLN) proprietary reaction databases public data commercial reaction databases U large number of reactions per year, strong medicinal chemistry bias wealth of reaction data extract some of the knowledge hidden in these data use this knowledge to assist the medicinal chemist suggest new, synthetically feasible molecules with desired bio profile

3. Reaction vectors From reaction database to knowledge base 1234 BondC-CC=OC-OHC-OR #00-22 reactant vector, R = (R1 + R2)product vector, P reaction vector, D = P - R 1234 BondC-CC=OC-OHC-OR #4120 1234 BondC-CC=OC-OHC-OR #4102 R1R2P Patel, H., Bodkin, M.J., Chen, B., Gillet, V.J.A Knowledge-Based Approach to De Novo Design Using Reaction Vectors, J. Chem. Inf. Model., 2009, ASAP article

4. From reaction vector to products (I) The reaction vector, D, equals the difference between the product vector, P, and the reactant vector, R D = P – R 1234 BondC-CC=OC-OHC-OR #4102 better descriptor is required Given a reaction vector, D, and a reactant vector, R, the product vector, P, can be obtained P = D + R Given a product vector, P, can we reconstruct the product molecule(s)?

5. Extended atom pairs atom typesatom pairs No.SymbolnprType 4C310C(3,1,0) 5O200O(2,0,0) 7C200C(2,0,0) Atom PairAtoms C(3,1,0)-2(1)-O(2,0,0)4-5 C(2,0,0)-2(1)-O(2,0,0)7-5 C(2,0,0)-3-C(3,1,0)2-4; 7-4 AP2: atoms 1 bond away AP3: atoms 2 bonds away n: number of bonds to heavy atoms p: number of π bonds r: number of ring memberships

6. From reaction vector to products (II) Atom PairCount C(1,0,0)-2(1)-C(2,0,0)2 C(2,0,0)-2(1)-C(2,0,0)1 C(2,0,0)-2(1)-C(3,1,0)1 C(3,1,0)-2(1)-O(2,0,0)1 C(3,1,0)-2(2)-O(1,1,0)1 C(2,0,0)-2(1)-O(2,0,0)1 C(1,0,0)-3-C(2,0,0)1 C(2,0,0)-3-C(3,1,0)2 C(2,0,0)-3-O(2,0,0)1 C(2,0,0)-3-O(1,1,0)1 O(2,0,0)-3-O(1,1,0)1 C(1,0,0)-3-O(2,0,0)1 C(2,1,0)-2(2)-O(1,1,0) C(3,1,0)-2(1)-O(2,0,0) wrong or missing atom pairs product vector (P = D + R) C(3,0,0)-2(1)-O(2,0,0) C(3,1,0)-2(1)-O(2,0,0)

7. Reaction vectors in action +1C(2,1,0)-2(2)-C(1,1,0)-2C(2,0,0)-2(1)-C(2,0,0) +1C(2,1,0)-2(1)-C(2,0,0)C(2,0,0)-2(1)-O(1,0,0) APs GainedAPs Lost New atoms/bonds added using APs gained Atoms/bonds selected for removal using APs lost Starting Molecule Reaction Vector Product Reaction

8. Advantages Does not require manual atom-atom mapping of the reaction centre Makes use of the synthetic chemistry data collected through the years Accounts for the synthetic accessibility of the proposed molecules – all transformations are derived from successful reactions Is fast to apply – no substructure searching is required

9. Good approach… so how is it… implemented?

10. Optimisation made easy build as an Eclipse plug-in => 100% Java

11. KNIME meets Chemaxon

12. Sketcher

13. File reader

14. Reaction generator

15. Convertor

16. Multi-objective ranking

17. File writer

18. Marvin Views

19. Looks great… but does it … work?

20. Reproducing reactions 5,695 diverse reactions create knowledge base 1 for each reaction 2 retrieve its reaction vector 3 apply the reaction vector to the starting materials 4 + -H 2 O is the product obtained in less than 30 seconds? 5 2,902 reaction vectors

21. How well did it work? Products generated for ~90% of the 5,695 reactions

22. How fast did it work? Median run time: 0.015 seconds per reaction

23. Epoxide reduction reproduced in large variety of environments (350 reactions) only one reaction was not reproduced Epoxide reduction

24. Works like a charm… More than 95% reproduced successfully epoxide reductionepoxide formationester to amide alcohol dehydration Friedel-Crafts acylation nitro reduction acid to aldehydenitrile to aldehyde nitrile hyrdrolysisalcohol amination aldol condensationalkene oxidation

25. Still works like a charm… More than 90% reproduced successfully olefin metathesisamide reductionether halogenation ozonolysis alkene halogenation Wittig-Horner Beckmann rearrangementClaisen rearrangement Dieckmann condensationolefination Robinson annulation

26. variety of environments were tested 79 out of 100 reactions were successfully reproduced 21% of the reactions were not reproduced mainly condensations (intra- and intermolecular) which result in ring closures Claisen condensation

27. Still works More than 50% reproduced successfully A large variety of reactions successfully reproduced Small difficulties with complex cycle formations improvements are on their way Cope rearrangement (67% success ) hetero Diels-Alder (73% success)Claisen condensation (79% success) Diels-Alder cycloaddition (49% success)Fischer indole synthesis (57% success)

28. Wow! Cool! It works! but what is its… use?

29. Generating new molecules Starting molecule Can the transform be applied? Apply reaction transform New molecule Select reaction transform Is a second reagent required? Select suitable reagent Discard reaction vector yes no Knowledge base Reagents database

30. rank the proposed new molecules direct the generation towards desired new molecules Multi-objective de novo design

31. Use case one: Lead optimisation Here is my starting material. What kind of (feasible) one step transformations may I make? starting molecule: Pencillin G An example from Patel, H., Bodkin, M.J., Chen, B., Gillet, V.J. A Knowledge-Based Approach to De Novo Design Using Reaction Vectors, J. Chem. Inf. Model., 2009, ASAP article

32. Lead optimisation (cntd.) An example from Patel, H., Bodkin, M.J., Chen, B., Gillet, V.J. A Knowledge-Based Approach to De Novo Design Using Reaction Vectors, J. Chem. Inf. Model., 2009, ASAP article Penicillin G

33. Use case two: Synthetic route I have this (active) fragment. Is there a route from it to the molecule I have in mind? reproducing known synthetic route – Plavix Synthetic route from Wang, L. et al., Synthetic Improvements in the Preparation of Clopidogrel, Org. Process Res. Dev., 2007, 11 (3), 487-489 An example from Patel, H., Bodkin, M.J., Chen, B., Gillet, V.J. A Knowledge-Based Approach to De Novo Design Using Reaction Vectors, J. Chem. Inf. Model., 2009, ASAP article Step No. applicable reaction vectors Total no. products generated 117158 211123 312124 441386 12 3 4

34. Use case three: Library design With which of these reagents will my starting material undergo reaction X? enumerate a library using a single reaction and a number of different reagents An example from Patel, H., Bodkin, M.J., Chen, B., Gillet, V.J. A Knowledge-Based Approach to De Novo Design Using Reaction Vectors, J. Chem. Inf. Model., 2009, ASAP article starting material reaction X (X = Suzuki coupling) 628 boronic acids as reagents

35. Library design (cntd.) 292 products generated

36. Summary The reaction vectors offer good way to explore the knowledge hidden inside reaction databases A variety of chemical reactions can be reproduced with this approach The method works fast The is applicable in different medicinal chemistry related scenarios The use of the method is made easy by variety of KNIME nodes which have been implemented

37. Acknowledgements Michael Bodkin for his continuous support both in and outside my daily work Hina Patel for creating the first prototype which sprung the reaction vectors into live (http://pubs.acs.org/doi/abs/10.1021/ci800413m) Dave Evans, Fred Ludlow, Swanand Gore, Dave Thorner, Maria Whatton, Juliette Pradon for many stimulating discussions and for their continuous support

38. Thank You! do you have any… questions, comments, recommendations?