Display options
Cite
Olivecrona M, Blaschke T, Engkvist O, et al. Molecular de-novo design through deep reinforcement learning. J Cheminform. 2017;9(1):48doi: 10.1186/s13321-017-0235-x.
Olivecrona, M., Blaschke, T., Engkvist, O., & Chen, H. (2017). Molecular de-novo design through deep reinforcement learning. Journal of cheminformatics, 9(1), 48. https://doi.org/10.1186/s13321-017-0235-x
Olivecrona, Marcus, et al. "Molecular de-novo design through deep reinforcement learning." Journal of cheminformatics vol. 9,1 (2017): 48. doi: https://doi.org/10.1186/s13321-017-0235-x
Olivecrona M, Blaschke T, Engkvist O, Chen H. Molecular de-novo design through deep reinforcement learning. J Cheminform. 2017 Sep 04;9(1):48. doi: 10.1186/s13321-017-0235-x. PMID: 29086083; PMCID: PMC5583141.
Copy Download .nbib Format: NLM
Share it on

J Cheminform. 2017 Sep 04;9(1):48. doi: 10.1186/s13321-017-0235-x.

Molecular de-novo design through deep reinforcement learning.

Journal of cheminformatics

Marcus Olivecrona, Thomas Blaschke, Ola Engkvist, Hongming Chen

Affiliations

  1. Hit Discovery, Discovery Sciences, Innovative Medicines and Early Development Biotech Unit, AstraZeneca R&D Gothenburg, 43183, Mölndal, Sweden. [email protected].
  2. Hit Discovery, Discovery Sciences, Innovative Medicines and Early Development Biotech Unit, AstraZeneca R&D Gothenburg, 43183, Mölndal, Sweden.

PMID: 29086083 PMCID: PMC5583141 DOI: 10.1186/s13321-017-0235-x

Abstract

This work introduces a method to tune a sequence-based generative model for molecular de novo design that through augmented episodic likelihood can learn to generate structures with certain specified desirable properties. We demonstrate how this model can execute a range of tasks such as generating analogues to a query structure and generating compounds predicted to be active against a biological target. As a proof of principle, the model is first trained to generate molecules that do not contain sulphur. As a second example, the model is trained to generate analogues to the drug Celecoxib, a technique that could be used for scaffold hopping or library expansion starting from a single molecule. Finally, when tuning the model towards generating compounds predicted to be active against the dopamine receptor type 2, the model generates structures of which more than 95% are predicted to be active, including experimentally confirmed actives that have not been included in either the generative model nor the activity prediction model. Graphical abstract .

Keywords: De novo design; Recurrent neural networks; Reinforcement learning

References

  1. J Mol Graph Model. 2004 Mar;22(4):263-73 - PubMed
  2. J Comput Aided Mol Des. 1992 Feb;6(1):61-78 - PubMed
  3. Nat Rev Drug Discov. 2005 Aug;4(8):649-63 - PubMed
  4. J Chem Inf Model. 2009 Jun;49(6):1514-24 - PubMed
  5. J Cheminform. 2009 Apr 28;1:4 - PubMed
  6. J Chem Inf Model. 2010 May 24;50(5):742-54 - PubMed
  7. Future Med Chem. 2011 Mar;3(4):415-24 - PubMed
  8. Nucleic Acids Res. 2012 Jan;40(Database issue):D1100-7 - PubMed
  9. Nat Chem. 2012 Jan 24;4(2):90-8 - PubMed
  10. PLoS Comput Biol. 2012;8(2):e1002380 - PubMed
  11. Nature. 2012 Dec 13;492(7428):215-20 - PubMed
  12. J Chem Inf Model. 2013 Jan 28;53(1):56-65 - PubMed
  13. J Chem Inf Model. 2013 Apr 22;53(4):783-90 - PubMed
  14. Mol Inform. 2013 Feb;32(2):133-138 - PubMed
  15. J Chem Inf Model. 2016 Feb 22;56(2):286-99 - PubMed
  16. Nature. 2016 Jan 28;529(7587):484-9 - PubMed
  17. J Cheminform. 2017 Mar 7;9:17 - PubMed
  18. ACS Cent Sci. 2018 Jan 24;4(1):120-131 - PubMed
  19. ACS Cent Sci. 2018 Feb 28;4(2):268-276 - PubMed
  20. J Chem Inf Comput Sci. 1994 Jan-Feb;34(1):207-17 - PubMed
  21. Neural Comput. 1997 Nov 15;9(8):1735-80 - PubMed

Publication Types