580 likes | 592 Views
Learn how to use referent tracking for building ontologies in the biomedical field. Explore the best practices and principles for ontology creation, as well as techniques for representing biomedical data and knowledge. The course includes hands-on exercises and student presentations.
E N D
Advanced Topics in Biomedical OntologyPHI 637 SEM / BMI 708 SEM Werner Ceusters and Barry Smith
Lecture 6Werner Ceusters Using Referent Tracking for building ontologies
Classes • Aug 31: Systems and techniques for representing biomedical data, information and knowledge in ontologies (WC) • Sept 7: Best practice principles for building domain ontologies, terms, and definitions (BS) • Sept 14: Basic Formal Ontology (BS) and the Ontology for General Medical Science (OGMS) • Sept 21: Introduction to the Protégé ontology editor and add-on tools (Neil Otte) • Sept 28: BFO, OGMS and the OBO Foundry (BS) • Oct 5: Using referent tracking for building ontologies (WC) • Oct 12: Team exercise: building an ontology (WC) • Oct 19: Team exercise: review of term-paper abstracts (WC, BS) • Oct 26: Principles for ontology change management in biomedical information systems (WC) • Nov 2: Ontological principles for combining healthcare data in big data repositories (WC,BS) • Nov 9: Team exercise: use OGMS to improve biomedical informatics resources (WC, BS) • Nov 16: Evaluation of ontologies (WC, BS) • Nov 30 and Dec 7: Student presentations.
Team exercise 1 • October 12: Building an ontology (WC) • class participants will be divided into groups. The task for each group will be: • to identify some area in which ontology methods can be of value in understanding issues related to patient well-being, along the lines illustrated in the pre-lecture readings. • to propose terms and definitions which need to be added (or linked) to OGMS to create a corresponding ontology. • to make the results available electronically by the end of class.
Pre-lecture reading test Arp R, Smith B, Spear AD. Building Ontologies with Basic Formal Ontology. MIT Press, 2015, chapter 7. Hogan WR and Ceusters W. Diagnosis, misdiagnosis, lucky guess, hearsay, and more: an ontological analysis. Journal of Biomedical Semantics 2016;7(54).
Q1. Definition of is_a (5%) • This BFO definition contains an implicit assumption. Which one?
Q2. (10%) • Give the formal definition for • C continuant_part_of D.
Q3. • Which of the following relational properties apply to located_in ? • Transitive • Symmetric • Reflexive • Antisymmetric • 5% for each correct property, -4% for wrong one, minimum: 0%.
Q4 (15%) • What is different for patient p1 in Fig.2 as compared to Fig.1? • Fig. 1 Fig. 2
Q5. Assertions can fail at the level of reference and at the level of compound expression. Fill out what is the case for each scenario.
Q1. Definition of is_a • This BFO definition contains an implicit assumption. Which one? • A and B are occurrent universals (5%)
Q2. • Give the formal definition for • C continuant_part_of D. • 10%
Q3 • Which of the following relational properties apply to located_in ? • Transitive (5%) • Symmetric (-4%) • Reflexive (5%) • Antisymmetric (-4%)
Q4 • What is different for patient p1 in Fig.2 as compared to Fig.1? in Fig.2 the diagnosis is wrong. (15%) • Fig. 1 Fig. 2
Q5. Assertions can fail at the level of reference and at the level of compound expression. Fill out what is the case for each scenario.
Representing specific entities explicitreferenceto the individualentities relevant to the accurate description of some portion of reality, ... Ceusters W, Smith B. Strategies for Referent Tracking in Electronic Health Records. J Biomed Inform. 2006 Jun;39(3):362-78.
Representing specific entities explicitreferenceto the individualentities relevant to the accurate description of some portion of reality, ... Ceusters W, Smith B. Strategies for Referent Tracking in Electronic Health Records. J Biomed Inform. 2006 Jun;39(3):362-78.
Portions of reality • entities (= particulars, = instances), • e.g.: me, my life; • relations, • e.g.: the 3-place parthood relation between me, my nose, and the temporal region at which it holds, • types, • universals, e.g. Nose, • defined classes, e.g. Crooked Nose, • configurations, • e.g. my nose now being part of me.
78 235 5678 321 322 666 427 Method: IUI assignment • Introduce anInstance Unique Identifier(IUI) foreach relevant particular (individual) entity Ceusters W, Smith B. Strategies for Referent Tracking in Electronic Health Records. J Biomed Inform. 2006 Jun;39(3):362-78.
Identifiers and pseudo-identifiers • Pseudo-identifiers: • ID-a: denotes nothing • ID-b: denotes ambiguously • Singularly unique identifier: ID-c • Non-singularly unique identifiers: ID-f, ID-d, ID-e
Reality representation https://www.statnews.com/2016/03/24/appendix-cancer-treatment/
Reality through representation https://www.statnews.com/2016/03/24/appendix-cancer-treatment/ this man this man #1 #1
Reality and representation https://www.statnews.com/2016/03/24/appendix-cancer-treatment/ this man this picture part this man #2 #1 #1 isAbout at t
Reality and representation this heart this picture part this heart #245 #12 #12 isAbout at t
Convention • Through x one sees x. • x stands proxy for x. • Alternative: ‘x’ stands proxy for x.
Configurations through Referent Tracking • When x or y is a continuant: x relation y at t • Otherwise: • x relationy • Where: • x is a particular, • relation (at) is a relation between x, y (and t), • y is a particular or a type, • t is a BFO:temporal_region, • x relation y at t is a configuration, • x relation y is a configuration. portions of reality
Referent tracking assertions • When x or y is a continuant: • Through: x relation y at t • One sees: x relation y at t • Otherwise: • Through: x relation y • One sees: x relation y
Examples of Referent Tracking assertions #1 participantOf #2at t1 #2 instanceOfDiagnosticProcess #1 participantOf#3 at t2 #3instanceOfLife #1 participantOf#4 at t3 #4instanceOfSurgicalProcedure #4 precededBy #2 #4 partOf #3 t3 partOf t2
‘#n’: (globally) singularly unique identifiers #1 participantOf #2at t1 #2 instanceOfDiagnosticProcess #1 participantOf#3 at t2 #3instanceOfLife #1 participantOf#4 at t3 #4instanceOfSurgicalProcedure #4 precededBy #2 #4 part of #3 t3 partOf t2
‘tn’: (globally) unique identifiers #1 participantOf #2at t1 #2 instanceOfDiagnosticProcess #1 participantOf#3 at t2 #3instanceOfLife #1 participantOf#4 at t3 #4instanceOfSurgicalProcedure #4 precededBy #2 #4 part of #3 t3 partOf t2
Identifier assignment • You can only assign an identifier to something existing. • However, besides the existence of the entity, what or how it is precisely does not need to be known.
Identifier assignment • You can only assign an identifier to something existing. • However, besides the existence of the entity, what or how it is precisely does not need to be known. #1 participantOf #3 at t2 #3 instanceOf Life
Identifier assignment • You can only assign an identifier to something existing. • However, besides the existence of the entity, what or how it is precisely does not need to be known. #1 participantOf #3 at t2 #3 instanceOf Life t2 can be used as ID for the temporal region during which #1 participates in his life even if nothing is known (yet) about the length of t2.
Do t1 and t2 denote distinct temporal regions? #1 participantOf #2at t1 #2 instanceOf Life #3participantOf#4 at t2 #4instanceOfLife #5instanceOfMonoZygoticTwinBrotherHood #5inheresIn#1 at t1 #5 inheresIn#3 at t2
Do t1 and t2 denote distinct temporal regions? #1 participantOf #2 at t1 #2 instanceOf Life #3 participantOf #4 at t2 #4 instanceOf Life #5 instanceOfMonoZygoticTwinBrotherHood #5 inheresIn #1 at t1 #5 inheresIn #3 at t2 • We will be able to tell when at least one of the twins dies: • If only one: no. • If both at the same time: yes.
Capacities for reasoners #1 participantOf #2at t1 #2 instanceOfDiagnosticProcess #1 participantOf#3 at t2 #3instanceOfLife #1 participantOf#4 at t3 #4instanceOfSurgicalProcedure #4 precededBy #2 #4 partOf #3 t3 partOf t2
Careful though ! #1 participantOf #2at t1 #2 instanceOfDiagnosticProcess #1 participantOf#3 at t2 #3instanceOfLife #1 participantOf#4 at t3 #4instanceOfSurgicalProcedure #4 precededBy #2 #4 partOf #3 t3 partOf t2 How does t1 relates to #2 ?
Careful though ! #1 participantOf #2at t1 #2 instanceOfDiagnosticProcess #1 participantOf#3 at t2 #3instanceOfLife #1 participantOf#4 at t3 #4instanceOfSurgicalProcedure #4 precededBy #2 #4 partOf #3 t3 partOf t2 How does t2 relates to #3 ?
Remember • Through: #1 partOf #2 at t1 • One sees: #1 partOf #2 at t1
Referent tracking assertions are real • Through: #1 partOf #2 at t1 • One sees: #1 partOf #2 at t1
Referent tracking assertions are real • Through: #1 partOf #2 at t1 • One sees: #1 partOf #2 at t1 • Thus we can assign identifiers to these assertions: • #3 stands proxy for #1 partOf #2 at t1 • And we can write: • #3 isAbout #1 at t3, • #3 isAbout#2 at t3, … • This is important for referent tracking systems which keep track of the faithfulness of its representations, but not for today’s exercise (and the assignment).
Use of Referent Tracking • Above all: • Representation of what is the case for particulars in some portion of reality: • Electronic healthcare systems • Gazetteers • Product or system maintenance systems. • But also: • As an aid to build application ontologies. • Forces you to think better about temporal aspects!
Time not visible anymore BFO-ontologies • C isaC1 = [def for continuants] for all c, t, if c instance_of C at t then c instance_of C1 at t. • C continuant_part_ofC1 = [def] for all c, t, if c instance_of C at t then there is some c1 such that c1 instance_of C1 at t and c continuant_part_ofc1 at t.
You must keep time in mind when crafting definitions! • ‘Persistent idiopathic facial pain (PIFP)’ = ‘persistent facial pain with varying presentations …’ persistent facial pain presentation type1 presentation type2 presentation type3 types t3 t2 t1 t1 t1 t2 t2 t3 t3 t1 t1 parti- culars t2 t1 t2 t2 t3 t3 t3 my pain her pain his pain
You must keep time in mind when crafting definitions! • ‘Persistent idiopathic facial pain (PIFP)’ = ‘persistent facial pain with varying presentations …’ • if the description is about types, then the threeparticularpains fall under PIFP. • if the description is about (arbitrary) particulars, then only her pain falls under PIFP.