1. Describing the Quality of Knowledge Contained in Biological and Medical Knowledge Bases Mor Peleg Department of Management Information Systems University of Haifa The Value of Information in Networked Contexts February 3, 2004

2. Knowledge Base (KB) KB - a centralized repository for information, stored in machine-readable format, usually on-line A KB system contains: –A knowledge model –Tools for information collection, organization, and retrieval

3. Components of a biological model Sequence components DB entries Cellular location Gene products Normal/mutated Biological process Proteolysis Transport Gene regulation Molecular function Clinical phenotypes

4. Goals Piece together biological data –Of various types, sources, and quality Develop a qualitative model at first –Data is noisy and incomplete Create a quantitative model eventually Store knowledge to allow –systematic evaluation by scientists –input for computer algorithms

5. Evaluation of other models We evaluated 11 models –from biology, business, and software eng. We wanted to combine the best aspects of two of the models Workflow Biomedical Biological Model + Concept Model  Process Model (TAMBIS+UMLS) Framework developed using Protégé-2000 Peleg, Yeh, and Altman, Bioinformatics, vol. 18, pp. 825-837, 2002

6. Mapping business workflow to biological systems Business Workflow model Biological Process Model Process model Structural modelOrganizational model Biomolecular complex (Replication complex) Biopolymer (Helicase) Role (DNA unwinding) member Organizational Unit (Medical School) Human Role (Dean) member Process model

7. Graphic dynamic & functional model Queries: All links supported only by speculation All processes inhibited by neuraminadase and occur in membrane All mutations that cause misfunctional processes and have clinical phenotype

9. Mapping to Petri Nets Explicitly represent states Verification of properties –liveness, boundedness Reasoning on dynamics –without t1, can we reach P4 from P1? Simulation Which of 2 models is correct? P1 t1 t4 P2 t2 P3 t3 P4 AND

10. Conclusion Our framework –pieces together qualitative biological data of various types, sources, and quality –supports identification of the quality of data in the KB »Useful for identifying areas for more experimentation »Useful to decide which conflicting facts are more credible –supports reasoning (queries) –enables verification of dynamic properties and simulation (prediction)

11. Representing levels of evidence in clinical guideline models

12. Thanks! morpeleg@mis.hevra.haifa.ac.il http://hevra.haifa.ac.il/mis/mor/

13. Do other models posses the desired properties? Model grafnestingstaticfunctiondynamicbio infoverifySimulati on tools Computatio nal model Query types* GO++-1,3 TAMBIS++DL1,3 EcoCyc++++frames1,2,3 Rzhetsky++++frames1,2,3 State- charts ++C+statechart OMT/UML++++C+/-statechart OPM++++I+/-Semi- formal PIF/PSL+IKIF BPML+CXML Workflow++++I++Petri Nets Petri Net++I In some ++Petri Nets DMD: 1,2,4,5 Meta: 2,3 our model++++I+++Petri Nets1-5 C= components, I = integrated * we considered models that can represent biological-specific information

14. Participant-Role Diagrams Individual molecule Complex Collection Functional role Disease role Participants Relations Roles role Complex-subunit Collection-participant Molecule-domain specialization Peleg, Gabashvili, and Altman. P IEEE 90(12): 1875-1886, 2002

15. Queries