Drug and Vaccine Discovery: Knowledge Graph + Apache Spark
0 likes•726 views
The document discusses the use of knowledge graphs and Apache Spark to enhance drug and vaccine discovery processes. It highlights the importance of data integration and querying capabilities to manage complex relationships within extensive datasets in R&D. The presentation outlines the architecture, querying methods, and future developments of the Bellman OSS SPARQL library, co-developed by GSK.
Drug and Vaccine Discovery: Knowledge Graph + Apache Spark
- 1. Ai Driven Drug and Vaccine Discovery: Knowledge Graph + Apache Spark John Hunter Lead Architect, Knowledge Graph, GSK
- 2. Agenda § Intro to drug R&D § Why a Knowledge Graph? § Knowledge Graph Data § Querying a Knowledge Graph § Bellman OSS Sparql library § The future
- 3. Intro to drug & vaccine R&D
- 4. Drug and Vaccine R&D pipeline • 12-15 years to bring a drug or vaccine to market • Phases • Research ⬅ Focus of this talk • Pre-clinical • Clinical trials • Government agency review • Post market monitoring
- 5. Typical R&D data use cases Accelerate research and development with data at scale • Discover • Encode • Predict
- 6. How to unify and manage siloed data for research?
- 8. Why a Knowledge Graph?
- 9. Why a Knowledge Graph? Connec@ons are priority constructs Disease Gene is_associated_with Subject Predicate Object
- 10. Why a Knowledge Graph? Data can be federated and is human + machine readable Compound Disease Gene Symptom Side Effect Pharma class Upregulates Downregulates Causes is_associated_with is_associated_with Treats Palliates Presents Includes
- 11. Why a Knowledge Graph on Spark? • Over 500 Billion connections and growing • High read/write throughput • Mature tech • Fantastic OSS and expert community
- 14. RDF Triples Ntriple format Disease Gene is_associated_with Subject Predicate Object <Disease> <is_associated_with> <Gene>
- 15. RDF Quads SPOG table <disease_5> <graph1> <disease_4> <equivalent_to> <associated_with> <gene3> <graph2> <disease_3> <disease_1> <gene1> <associated_with> O : String G : String <disease_2> <gene2> <associated_with> S : String <graph2> P : String <graph1> …
- 16. RDF data store Data architecture <disease_5> <graph1> <disease_4> <equivalent_to> <associated_with> <gene3> <graph2> <disease_3> <disease_1> <gene1> <associated_with> O : String G : String <disease_2> <gene2> <associated_with> S : String <graph2> P : String <graph1> graph1 <associated_with> Part001.parquet Part002.parquet … PartNNN.parquet <equavalent_to> Part001.parquet Part002.parquet … PartNNN.parquet graph2 … …
- 17. RDF data store Addi@onal requirements • Data abstraction • ACID transactions: Atomic writes, isolated reads • Dedup • Incremental queries • Apache Hudi, Delta Lake
- 19. Sparql Sparql Protocol and RDF Query Language SELECT ?s ?p ?o FROM <graph1> FROM <graph2> WHERE { ?s ?p ?o } LIMIT 10
- 20. Sparql • Commercial offerings • OSS offerings • Apache Jena • Blazegraph • Sansa Stack Build vs Buy
- 21. Sparql Bellman Engine • OSS project sponsored by GSK • Co-developed by GSK & 47Degrees • Please star and contribute! • https://github.com/gsk-aiops/bellman
- 22. Sparql • Parser • Compiler • Static analysis • Optimizer • Engine • Formatter Bellman architecture
- 23. Demo • h[ps://github.com/gsk-aiops/bellman-tools Querying Knowledge Graphs using Bellman
- 24. Future • Complete Sparql language coverage • Optimization of queries • Optimization of data on disk • Incremental / point in time queries Bellman next steps
- 25. Thank you Links: https://github.com/gsk-aiops/bellman https://github.com/gsk-aiops/bellman-tools Connect on Github: https://github.com/JNKHunter