Presentation is loading. Please wait.

Presentation is loading. Please wait.

Semantics and Literature 1 st HCLSIG Meeting Cambridge 25-26 January 2006 Davide Zaccagnini MD, MS.

Similar presentations


Presentation on theme: "Semantics and Literature 1 st HCLSIG Meeting Cambridge 25-26 January 2006 Davide Zaccagnini MD, MS."— Presentation transcript:

1 Semantics and Literature 1 st HCLSIG Meeting Cambridge 25-26 January 2006 Davide Zaccagnini MD, MS

2 Technology Source NLPText SemanticsSchemas DB Integration Data Bases Semantics and Literature ‘semanticizing’ is possible, but only part of the solution Life scientists need Querying Browsing Visualizing

3 NLP, DB integration 2 1 INPUT Free text Semantics 3 text strucure: paragraphes sentences verbs words Semantics and Literature: end to end solution 4 Knowledge Base Queries and visualization Knowledge discovery D ependencies concepts dependencies relationships

4 NLP: extracting the information –Strucutral and grammatical parsing Paragraphes, sentences, words, negation, modality (uncertainty) –Syntactical parsing Dependencies between terms –Ontology-based semantic tagging Infers concepts and relationships

5 Syntactics and Semantics: creating the ontology from text “The binding of VEGF to its receptor on a vascular endothelial cell activates a variety of signaling molecules.” “The binding of VEGF to its receptor on a vascular endothelial cell activates a variety of signaling molecules.” “The binding of VEGF to its receptor on a vascular endothelial cell activates a variety of signaling molecules.” Example Semantic tagger

6 Leveraging information Semantic queries –Closed queries –Open queries Graphical view of query result(s) Backtrack concepts to documents

7 Use case The problem –Anti-diabetic drug at advanced development stage –Unexpected results in clinical trials –Need to understand why The approach To semantically integrate –8800 abstracts (diabetes, PKC, QT prolongation) –Swiss Prot and proprietary data-bases –Gene Ontology To allow –Querying –Visualization

8 Closed query: what paths connect ‘PKC Beta inhibitor’ and ‘Diabetes’ Paths Connecting path via concept ’ADIPOCYTE’ Paths are classified by links directions (Incoming/Outgoing) Both Out include IS_As (common ancestors) One In/One Out include non-IS_As (i.e., X affects Y)

9 Graphical representation of selected paths between concepts ‘Adipocyte path’ intersection

10 Open query: What are the physiological roles of PKC? Approximately 1600 relations are found

11 Backtrack from facts to documents.

12 Plausible pathways Nitric oxide AGE’s High glucose Hypoxia PKC activation VEGF secretion Nifedipine (calcium channel blocker) Detrimental: Causes arterial hypertension Causes Retinal vascular (hyper)permeability Causes retinal angiogenesis Beneficial: Protects cardiac microcirculation Improves cardiac performance Stimulates decay-accelerating factor B2 receptor Might be specific for: Smooth muscle cells & podocytes Use case: What are the physiological roles of PKC?

13 Conclusions: ROI ‘New’ biological knowledge discovered (article in print) Dramatic reduction in drug development time (2 days for understanding what was going on) Moving forward to integrate clinical notes from trials and micro-arrays data

14 Thank you Davide Zaccagnini davide@landcglobal.com Language and Computing www.landcglobal.com

15

16

17 Approximately 1600 relations are found We can refine our search by filtering on specific relations Use case: What are the physiological roles of PKC?

18 Backtrack from facts to documents.

19 Browse for the knowledgename representation: Search on terms synonyms Use case: What are the physiological roles of PKC?

20 Drag-Drop the domain concept ‘Protein kinase C’ in the onelinktoofartree-bean Use case: What are the physiological roles of PKC?

21

22 Browse the relations for interesting leads We take VEGF as example case Nifedipine -> B2 receptor -> PKC activation -> VEGF secretion Build up a pathway by Examining relations for identified leads: e.g. vascular endothelial growth factor as input for the onelinktoofartree Use case: What are the physiological roles of PKC?

23 VEGF: filter: localisation related links Apparent relations: Expressed in: podocytes, smooth muscle cells & retinal endothelial cells Expressed in: high glucose/early diabetes elevation of VEGF in diabetic patients Use case: What are the physiological roles of PKC?

24 VEGF: filter: positively regulates links Confirmed relations: Expressed by smooth muscle cells/stimulated by nifedipine Expressed in high glucose Expressed under hypoxia PKC activation -> VEGF Use case: What are the physiological roles of PKC?

25 NO (nitric oxide) -> VEGF AGE’s -> VEGF VEGF: filter: mediates Detrimental relations: Causes arterial hypertension Causes Retinal vascular (hyper)permeability Causes retinal angiogenesis e.g. Vascular complications Beneficial relations: VEGF -> decay accelerating factor Use case: What are the physiological roles of PKC?

26 Protective role cardiac microcirculation Improvement cardiac performance in diabetes VEGF: filter: Role relation Use case: What are the physiological roles of PKC?

27 Plausible pathways Nitric oxide AGE’s High glucose Hypoxia PKC activation VEGF secretion Nifedipine (calcium channel blocker) Detrimental: Causes arterial hypertension Causes Retinal vascular (hyper)permeability Causes retinal angiogenesis Beneficial: Protects cardiac microcirculation Improves cardiac performance Stimulates decay-accelerating factor B2 receptor? Might be specific for: Smooth muscle cells & podocytes Use case: What are the physiological roles of PKC?

28

29 Benefits of using L&C technology in KD solution. Full parser: Full parsing analyzes full sentences, it is the only way to derive correct and detailed extractions. Semantic information is used –to disambiguate possible meanings of a term or a relation. –to restrict the possible parser outcomes, improving the precision of the extracted information –... The system learns: created KB concepts are possible candidates for ontology maturation. A powerfull query engine –that can derive any piece of information in a limited amount of time (scalability), –that is the basis of different search methodologies (now only two of them are implemented, the open and closed query). Every piece of information –can be traced to its source and validated –can be assembled in a connecting network Information is shown in a hyperbolic graph, a very effective way to show complex networks.

30 Information in HCLF –Available in large amounts, but in disparate forms –Electronic medical libraries are improving, but are vastly pre-semantic, text search- based –Not only literature, but data-bases, taxonomies, local ontologies

31 The KD platform: DB integration MaDBOks ® Diseases Disease X Proteins Abnormal Protein Database Table Column Protein 1 Protein 2 Protein 3 Protein 4 Protein 5 Protein 6 Ontology MadBOKS Is a pathologic marker of Is a SQLa

32 Language B The KD platform: Medical Ontology LinKBase®: LinKBase KB A Medical Semantic Network Language A Lexic on Gram mar Proprietary Decision Support or E&M Coding Rules MEDCIN SNOMED CT ICD-9 & ICD-10 MedDRA

33 Semantics and Literature Outline –The information problem in health care and life sciences –Technologies and processes –Use case, demo

34 Accessing the information: closed queries X P Q R S T Y B N G B N G 3 iterations this can (needs to) be configured Paths are classified by links directions (Incoming/Outgoing) Both Out include IS_A (common ancestors) One In/One Out include non-IS_A (i.e., X affects Y)

35 The linktype tree bean can be used to browse available relations - relationgroups Relation types extracted from text are specified under ‘onto-creation linktype’ Relations are organized hierarchically As example we will filter the Onelinktoofartree on Relations of the kind ‘positively regulates’ only Use case: What are the physiological roles of PKC?

36 Protein kinase C Examine relations related to PKC Use filter to refine relations Examine relations related to VEGF Use filter to refine relations Filters Role:Role relation Localisation:Localisation related link Regulation:Negatively regulates Positively regulates Preclusive relation Other:Affects, mediates, temporal relation, direct relation,... VEGF potential target Cardiac performance


Download ppt "Semantics and Literature 1 st HCLSIG Meeting Cambridge 25-26 January 2006 Davide Zaccagnini MD, MS."

Similar presentations


Ads by Google