Knowledge Graphs: Changing How We Think About Data
Download as PPTX, PDF5 likes793 views
The document discusses a collaborative conversation between PHUSE, Stardog, and the FDA about knowledge graphs and their application in transforming data handling in clinical research and regulatory compliance. It outlines various linked data projects and the importance of using Resource Description Framework (RDF) to foster a standardized understanding of data across industries. The conversation emphasizes the need for shared terminology, ontology-driven standards, and real-world entity representation to facilitate better data management and accessibility.
![• Knowledge Graphs [ ~20 min]
• Cross Industry Perspective [~10min]
• Open Discussion [~30min]
Outline
Most important: “A Conversation...”
3](https://image.slidesharecdn.com/knowledgegraphschanginghowwethinkaboutdata-tw-200817185706/85/Knowledge-Graphs-Changing-How-We-Think-About-Data-3-320.jpg)
Knowledge Graphs: Changing How We Think About Data
- 1. A Conversation with PHUSE, Stardog, and FDA DRSI Knowledge Graphs: Changing How We Think About Data.
- 2. Presenters Tim Williams • Knowledge Graph Project Lead, PHUSE • Lead Statistical Solutions Analyst, UCB Biosciences Laura Firey • Product Manager, Stardog 2
- 3. • Knowledge Graphs [ ~20 min] • Cross Industry Perspective [~10min] • Open Discussion [~30min] Outline Most important: “A Conversation...” 3
- 4. PHUSE Linked Data Projects • CDISC Foundational Standards in RDF • CDISC Conformance Checks • Reusing Medical Summaries for Enabling Clinical Research • Regulatory Guidance in RDF • Clinical Program Design in RDF • CDISC Protocol Representation Model in RDF • Analysis Results & Metadata • RDF Data Cubes for clinical trial results • Clinical Trials Data as RDF • Study Data Tabulation Model as Linked Data • Going Translational with Linked Data • Study Data Validation and Submission Conformance • Pre-clinical data + submission metadata 4
- 5. PHUSE Linked Data Workshop 5
- 7. Knowledge Graph Data-Centric Model • Use-case neutral • Real-world processes, entities, relationships Person 1 Hookah Person 2 • Knowledge Graph as Resource Description Framework (RDF) e-Cigarette smokes e-Cigarette smokes smokes smokes The whiteboard model you draw is the data. 7
- 8. • Shared Definitions and Understanding – What is a Tobacco Product? – Who is a Tobacco User? • Coding of – Medical Conditions – Adverse Events – Products, Manufacturers • Data Classification, Rules, Validation.. • Ontology Driven/Supported Knowledge Graphs Facilitate Standards 8
- 9. • Dictionary – terms and their definitions • Taxonomy – class hierarchy • Thesaurus – relationships between terms Ontology-based Knowledge Graph Standards • Rules and Restrictions – Group membership, exclusions, types – Employ reasoner, infer values, relations A Data Engineer’s Guide to Semantic Modelling Ilaria Maresi (June 2020) http://blog.thehyve.nl/news/ebook-semantic-model What is an Ontology? 9
- 10. Standards: Identifier Examples 1,2,3-TrihydroxypropaneGlycerolGlycerin http://dbpedia.org/page/GlycerolGlycerin • Common terminology • Link to other Knowledge Graphs • Manufacturers• Devices • Products • Ingredients RDF • Uniform Resource Identifier (URI) • Internationalized Resource Identifier (IRI) 10
- 11. • What ingredient may have contributed to Pam having COPD while Ray does not? Knowledge Graph: Easy Answers to Complex Questions COPD 11
- 12. Ray_GillettePam_Poovey COPD Hookah smokes smokes smokes smokes e-Cigarette e-Cigarette :man_v241kz Hookah_Me_Up! Tobacco :ing_a6766s Dried_Mint :ing_2we3q4 Honey Glycerin :ing_xr7234:ing_3j6g2i manufacturer:pro_81fq21 contains ingredient Mint_Shisha Mint_Flavor :ing_hjk66g Nicotine :ing_q77e3u:man_5ez82i e-CignatureMango_Flavor :ing_4fg421 Nicotine :ing_q77e3u contains contains : pro_0wq01h:pro_a60e2b ingredientingredient manufacturer Mango_Flavor_Pod Mint_Flavor_Pod :dev_zr10q2 :dev_u18lsd :dev_aq174h :per_452h78:per_xp52kl sex age wt sex age wt 12
- 13. Exposure Data View the interactive visualization at: https://bit.ly/PamAndRay Data error Person Demographics Entity Type Device Ingredients Product Manufacturer 13
- 14. This seems complicated! “People think RDF is a pain because it is complicated. The truth is even worse. RDF is painfully simplistic, but it allows you to work with real-world data and problems that are horribly complicated.” - attributed to Dan Brickley and Libby Miller 14
- 15. Data https://bit.ly/PamAndRayTTL View the data file at: 15
- 16. Pam’s Unique Exposure Ingredient? PamUniqueIngred.rq 16
- 17. • Who is a TobaccoUser ? Ontology Revisited • uses • smokes • consumes • chews ... User Tobacco • What is a TobaccoProduct? Tobacco Hookah Cigar e-Cigarette Cigarette contains contains contains ? ? Mint_Shisha Corojo Virginia Mango_Flavor_Pod contains ingredient ingredient ingredient ingredient contains ingredient :Person :Device :Device :Product :Ingredient :Product :Ingredient 17
- 18. Ray_Gillette :per_452h78 Pam_Poovey :per_xp52kl :dev_zr10q2 :man_v241kz :dev_u18lsd Hookah_Me_Up! :dev_aq174h Hookah Tobacco :ing_a6766s Dried_Mint :ing_2we3q4 Honey Glycerin :ing_xr7234:ing_3j6g2i smokes contains contains manufacturer : pro_0wq01h:pro_a60e2b smokes smokes smokes :pro_81fq21 contains e-Cigarette e-Cigarette Mint_Shisha Mango_Flavor_Pod Mint_Flavor_Pod sex age wt sex age wt ingredient Tobacco Smoker: The path between Person and Ingredient :Ingredient :Product :Device:Person :Person 18
- 20. Tobacco Smoker: Ontology Definition & Query TobaccoSmoker-Infer.rq 20
- 21. Data Validation Validating Ray’s Demographics data 21
- 22. SHApes Constraint Language (SHACL) “Person Shape” (Validation Constraints) Person Data Data Has Shape. Validation has Shape. 22
- 23. SEX AGE M SHACL Person Shape WEIGHT Demographics sex: M age: 43 wt: 165 Validating Ray’s Demographics Ray Gillette Data Entry M/F/U 165 403 > 0 < 800 > 0 < 120 Violation! 23
- 24. Validating Age with SHACL https://bit.ly/RayDemogAndSHACL View the interactive visualization at: Instance Data SHACL Shape 24
- 25. Demographics, Constraints, Report https://bit.ly/RaySHACLResult View the interactive visualization at: Instance Data SHACL Shape Validation Result 25
- 26. Metadata is Part of the Graph Ray_Gillette M 165 sex age wt 2020-08-01 Duke U. Clinic Dr. A. Krieger enteredBy site entryDate :per_g367k1 403:per_452h78 26
- 27. FAIR Data https://www.go-fair.org/fair-principles/ • Ontologies Mapping https://www.pistoiaalliance.org/projects/current-projects/ontologies-mapping/ • FAIR Implementation Project & Toolkit https://www.pistoiaalliance.org/projects/current-projects/fair-implementation/ Findable Accessible Interoperable Reusable FAIR Data is Linked Data is a Knowledge Graph 27
- 28. Knowledge Graphs : Now for a cross-industry perspective 28
- 29. Additional Reading General / Introductory • A Data Engineer’s Guide to Semantic Modeling - Maresi . Free e-book download. • The Data Centric Revolution - McComb Technical • Semantic Web for the Working Ontologist (3rd Ed, 2020) – Hendler, Gandon, Allemang • Demystifying OWL for the Enterprise – Uschold, Ding, Groth • Validating RDF Data – Gayo, Prud’hommeaux, Boneva . Comparison of SHEX and SHACL. • Learning SPARQL - DuCharme . Learn RDF by querying the data. • 3D-force-graph Network graph visualizations in this presentation. https://bit.ly/PamAndRay https://bit.ly/RayDemogAndSHACL https://bit.ly/RaySHACLResult https://bit.ly/PamAndRayTTL Pam and Ray Exposure Data 29
- 31. The Roofshot / Moonshot Manifesto Concept & Image Attribution: https://rework.withgoogle.com/blog/the-roofshot-manifesto/ Examples 1. Unique Identifiers 2. Validation Rules in SHACL 3. Open Ontology Development for your domain Invent and apply state- of-the-art Roofshot Incremental impacts in production Moonshot • Enterprise and Industry Knowledge Graphs • Across the Data Life Cycle 31
- 32. Industry Knowledge Graphs Demonstrations Results Data as RDF Enterprise Knowledge Graphs Prototype Industry Standards & Models Terminology and Coding The Stairway to the Stars Manifesto 32 Unique Identifiers for Pharma Validation Rules in SHACL Study Design Study Protocol