A Knowledge Graph for Reaction & Synthesis Prediction (AstraZeneca)
1 like528 views
The document summarizes the development of a reaction knowledge graph (RKG) by AstraZeneca to predict novel chemical reactions and assist in synthesis planning. Key points include: 1) The RKG integrates reaction and molecule data from multiple sources and enriches the data with identifiers, templates, and other metadata. 2) Graph analytics and machine learning models are used to provide insights into reaction patterns and predict new reactions and syntheses. 3) Future work includes expanding the RKG with AstraZeneca's full collection and developing link prediction workflows to further support computer-aided synthesis prediction.
![Synthesis Trees
MATCH p=(e:Molecule)<-[:IS_PRODUCT|IS_REACTANT*2..10]-(s:Molecule)
WHERE s <> e
AND e.inchikey = $inchikey
WITH p, nodes(p) as path_nodes, size(relationships(p)) as path_length
WHERE path_length <= $max_depth AND size(path_nodes) =
size(apoc.coll.toSet(path_nodes))
RETURN p, size(apoc.coll.toSet(path_nodes)) as num_nodes, path_length
ORDER BY path_length asc
LIMIT $limit_val
20
Reaction
Knowledge
Graph
Chemistry
aware
Backend](https://image.slidesharecdn.com/astrazenecareactionknowledgegraph-neo4jconnections-221213014015-91026a95/85/A-Knowledge-Graph-for-Reaction-Synthesis-Prediction-AstraZeneca-20-320.jpg)
![Link Prediction Using Molecule & Reaction Templates
Flaticon.com
MATCH (RV:ReactionVariation {source_name: ‘USPTO'}),
(R:Reaction)-[:HAS_VARIATION]->(RV),
(RT:ReactionTemplate {radius_1:True})<-[:HAS_FORWARD_TEMPLATE]-(R),
(RT)<-[IRT1:IS_REACTANT_TEMPLATE]-(MT1:MoleculeTemplate {radius_1:True})-[reacts:REACTS]-
(MT2:MoleculeTemplate {radius_1:True})-[IRT2:IS_REACTANT_TEMPLATE]->(RT)
WHERE MT1 <> MT2
RETURN RT, IRT1, MT1, reacts, MT2, IRT2
21
Link Prediction
Models](https://image.slidesharecdn.com/astrazenecareactionknowledgegraph-neo4jconnections-221213014015-91026a95/85/A-Knowledge-Graph-for-Reaction-Synthesis-Prediction-AstraZeneca-21-320.jpg)
![Link Prediction Using Molecules & Reactions
Flaticon.com
MATCH (rv:ReactionVariation {source_name: ‘USPTO’})-->(rc:ReactionClassification)
WHERE NOT rc.classification_id STARTS WITH ‘0’
WITH rv
MATCH (r:Reaction)-->(rv)
MATCH (rm:Molecule)-[rr:IS_REACTANT]->(r)-[pr:IS_PRODUCT]->(pm:Molecule)
RETURN rm, rr, r, pr, pm
22
Link Prediction
Models](https://image.slidesharecdn.com/astrazenecareactionknowledgegraph-neo4jconnections-221213014015-91026a95/85/A-Knowledge-Graph-for-Reaction-Synthesis-Prediction-AstraZeneca-22-320.jpg)
Ad
More Related Content
What's hot (20)
Similar to A Knowledge Graph for Reaction & Synthesis Prediction (AstraZeneca) (20)
Ad
More from Neo4j (20)
Ad
Recently uploaded (20)
A Knowledge Graph for Reaction & Synthesis Prediction (AstraZeneca)
- 1. A Knowledge Graph for Reaction & Synthesis Prediction Christos Kannas 1, Tomas Bastys 1, Thierry Kogej 1 1 MolecularAI, Discovery Sciences, R&D, AstraZeneca, Gothenburg, Sweden
- 2. Outline 1 Background 2 Building the graph 3 Using the graph 4 Future Work 5 Conclusion 2
- 3. Background 1
- 4. Drug Discovery Process Design • Develop • Evaluate Make • Prioritise Synthesis • Synthesise Test • Run Assays • Compound Selection Analyse • Data Analysis • Hypothesis 4
- 5. Objectives Predict Novel Reactions Insights about Reactions Insights about Reactants, Reagents, Products Assist Synthesis Prediction 5
- 6. Reaction Data Sources (1) Lowe, D. Chemical Reactions from US Patents (1976-Sep2016), 2017. https://doi.org/10.6084/m9.figshare.5104873.v1. (2) Kearnes, S. M.; Maser, M. R.; Wleklinski, M.; Kast, A.; Doyle, A. G.; Dreher, S. D.; Hawkins, J. M.; Jensen, K. F.; Coley, C. W. The Open Reaction Database. J. Am. Chem. Soc. 2021, 143 (45), 18820–18826. https://doi.org/10.1021/jacs.1c09820. AZ Reaction Data USPTO Reactions 1 Open Reaction Database 2 US Patents & Trademark Office 6
- 7. Reaction Record Example USPTO Identifier: US20120270775A1;94;2012 • Reaction Identifier • Source Name • Date • Reaction Yield Reactants Reagents Products • Classification Id • Classification Name Reaction Variation Reaction Reaction Classification Molecules 7
- 8. Reaction Knowledge Graph – Graph Model 8
- 9. Reaction Knowledge Graph – Building the graph 2
- 10. Reaction Knowledge Graph Chemistry aware Backend AZ Reactions AZ Molecules & Building Blocks Public Reaction Data Reaction Knowledge Graph – Architecture 10
- 11. Reaction Data ETL pipeline Extract • Get data from flat files Transform • Validate & select reaction records • Prepare data for loading Load • Generate Nodes & Relationships • Store Nodes & Relationships to Reaction Graph DB 11
- 12. Reaction Knowledge Graph – Graph Model 12
- 13. Molecule Enrichment Pipelines • Compound Identifiers • Building Block Identifiers Compound Identifiers • Update in stock status of molecules In stock status 13
- 14. Reaction Knowledge Graph – Graph Model 14
- 15. Template Enrichment Pipeline Process • Generate Reaction Templates • Binary Reactions • Radius Range Store • Store Reaction, Molecule Template Nodes & Relationships (1) Kannas, C.; Thakkar, A.; Bjerrum, E.; Genheden, S. Rxnutils – A Cheminformatics Python Library for Manipulating Chemical Reaction Data. 2022. https://doi.org/10.26434/chemrxiv-2022-wt440-v2. (2) Thakkar, A.; Kogej, T.; Reymond, J.-L.; Engkvist, O.; Bjerrum, E. J. Datasets and Their Influence on the Development of Computer Assisted Synthesis Planning Tools in the Pharmaceutical Domain. Chem. Sci. 2019, 11 (1), 154–168. https://doi.org/10.1039/c9sc04944d. (3) Coley, C. W.; Green, W. H.; Jensen, K. F. RDChiral: An RDKit Wrapper for Handling Stereochemistry in Retrosynthetic Template Extraction and Application. J. Chem. Inf. Model. 2019, 59 (6), 2529–2537. https://doi.org/10.1021/acs.jcim.9b00286. 15
- 16. Reaction & Molecule Templates 16
- 17. Reaction Knowledge Graph – Graph Model 17
- 18. Reaction Knowledge Graph – Using the graph 3
- 19. Insights - Graph Analytics Claudio Rocchini, 2017, https://commons.wikimedia.org/wiki/File:Graph_betweenness.svg https://mathematica.stackexchange.com/questions/31962/how-can-i-highlight-and-individually-color-the-connected-components-of-a-graph https://neo4j.com/blog/graph-algorithms-neo4j-weakly-connected-components/ • How is RKG structured • Important: • Reactions (maximize accessible chemical space) • Molecules (most needed chemicals) An undirected graph colored based on the betweenness centrality of each vertex from least (red) to greatest (blue). 19 Strongly connected components, each component is shown by a different colour. Weekly connected components, each component is shown by a different colour.
- 20. Synthesis Trees MATCH p=(e:Molecule)<-[:IS_PRODUCT|IS_REACTANT*2..10]-(s:Molecule) WHERE s <> e AND e.inchikey = $inchikey WITH p, nodes(p) as path_nodes, size(relationships(p)) as path_length WHERE path_length <= $max_depth AND size(path_nodes) = size(apoc.coll.toSet(path_nodes)) RETURN p, size(apoc.coll.toSet(path_nodes)) as num_nodes, path_length ORDER BY path_length asc LIMIT $limit_val 20 Reaction Knowledge Graph Chemistry aware Backend
- 21. Link Prediction Using Molecule & Reaction Templates Flaticon.com MATCH (RV:ReactionVariation {source_name: ‘USPTO'}), (R:Reaction)-[:HAS_VARIATION]->(RV), (RT:ReactionTemplate {radius_1:True})<-[:HAS_FORWARD_TEMPLATE]-(R), (RT)<-[IRT1:IS_REACTANT_TEMPLATE]-(MT1:MoleculeTemplate {radius_1:True})-[reacts:REACTS]- (MT2:MoleculeTemplate {radius_1:True})-[IRT2:IS_REACTANT_TEMPLATE]->(RT) WHERE MT1 <> MT2 RETURN RT, IRT1, MT1, reacts, MT2, IRT2 21 Link Prediction Models
- 22. Link Prediction Using Molecules & Reactions Flaticon.com MATCH (rv:ReactionVariation {source_name: ‘USPTO’})-->(rc:ReactionClassification) WHERE NOT rc.classification_id STARTS WITH ‘0’ WITH rv MATCH (r:Reaction)-->(rv) MATCH (rm:Molecule)-[rr:IS_REACTANT]->(r)-[pr:IS_PRODUCT]->(pm:Molecule) RETURN rm, rr, r, pr, pm 22 Link Prediction Models
- 23. Future Work 4
- 24. Reaction Knowledge Graph Chemistry aware Backend Analytics on Full AZ Collection Graph Data Science 24
- 25. Link Prediction Workflows 2 5 1 4 7 6 Can nodes 1 and 2 form link? Probability p of link If p above cut-off: 2 1 + P Product template construction TP Molecule Template Space Molecule Space Transformer for Product Prediction ML for Link Prediction 25
- 26. Reaction Knowledge Graph – Computer-Aided Zynthesis Prediction Ecosystem Integration Reaction Knowledge Graph Chemistry aware Backend CAZP Ecosystem 26
- 27. Conclusion Explore Chemical Reaction Space Exploit Chemical Reaction Space Assist Synthesis & Reaction Prediction 27
- 28. Acknowledgements Molecular AI • Tomas Bastys • Emma Rydholm • Emma Svensson • Thierry Kogej • Samuel Genheden • Ola Engkvist R&D IT • AJ Vignesh • Rathi Prakash • Alla Bushoy 28
- 29. Thank you. 29
- 30. Confidentiality Notice This file is private and may contain confidential and proprietary information. If you have received this file in error, please notify us and remove it from your system and note that you must not copy, distribute or take any action in reliance on it. Any unauthorized use or disclosure of the contents of this file is not permitted and may be unlawful. AstraZeneca PLC, 1 Francis Crick Avenue, Cambridge Biomedical Campus, Cambridge, CB2 0AA, UK, T: +44(0)203 749 5000, www.astrazeneca.com 30