Exercise #4: Building and Using an OWL Ontology (in Protégé)

Due Tues, March 1st, 9am

For exercise 2, I demonstrated how a simply rule-based system might encode some cancer staging rules. For this exercise, we'll take a closer look at the same domain: The official AJCC staging rules for non-small cell Lung Cancer. These staging rules are presented on the cancer.gov website . However, the precise details of these complex rules are not always clear; it may be useful to compare multiple versions - e.g., see http://www.health-alliance.com/Cancer/Lung/staging.html.

As I showed in class, the general idea is to use the automatic subsumption capability of a description logic reasoning system to classify examples of specific tumors and their metastases into one of seven AJCC stages, from Stage 1a to Stage 4. Thus, for this assignment you will need to use the OWL plug-in (this is included in the "full" Protege 3.0 installation), and the RACER classification engine. You need to download, install and start the RACER program separately from Protege: the download pages from RACER are very straightforward.

In a separate email, I have sent you all a copy of the Protege project and OWL file for "tnm-lungAblated" -- This project is incomplete, and your job is to restore it so that all of the "specific tumor examples" can be successfully and appropriately classified into the AJCC stages.

More precisely, you must define additional types of the class "lungTumor"; the version I gave has definitions for N0, T1, T2, M0 and M1 -- you must provide definitions for N1, N2, N3, T3, T4. Likewise, the ablated version has a class "StagedLungCancer" with Stage1a and Stage1b defined; you must add definitions of Stages 2a, 2b, 3a, 3b, and 4. Finally, all of the "SpecificTumorExamples" should be successfully classified into their appropriate stages. With the ablated version, only one of these will be classified by RACER: the class "SmallUpperLeftLobeTumourWithoutMetastasis" will be a subclass of Stage1a (there are no examples of Stage1b in my list). Remember the "classify taxomony" button will not work unless you have RACER running concurrent with Protege.

A few words about the oncology: The primary difference between N2 and N3 is contralateral versus ipsalateral nodal metastases -- i.e. whether the involved node is in the same lung (ipsalateral) or in the opposite lung (contralateral) from the primary tumor. Also, the AJCC wording for N1 is atrocious: " Metastasis to ipsilateral peribronchial and/or ipsilateral hilar lymph nodes, and intrapulmonary nodes . . . ". Please interprete the above to mean: Metastases to ipsalateral {peribrochial, or hilar, or intrapulmonary} nodes. (Basically, I think that all uses of the word "and" in this AJCC sentence are misleading and incorrect.)

In the "ablated" ontology I sent to you, you will find useful concepts under the class Metastasis that define things like "regionalLymphNodes" as well as contralateral and ipsalateral. These definitions may not be the best, and you should feel free to ignore or modify these as you see fit. In addition, depending on how you define the missing classes and stages, you may want to modify the specificTumorExamples. (Of course, my solution to this assignment uses the examples exactly as they are. However, I recognize that different ontologers may wish to model the world differently... )

Credit goes to Olivier Dameron, post-doc at Stanford University and OWL guru, for creating the original TNM ontology. Additional examples of OWL ontologies can be found at http://www.owl-ontologies.com/index.html. These are mostly non-medical, although there are some bio examples, such as Tambis and BioPax.

Deliverables:

1. A Protégé OWL ontology (Just the two files -- .pprj and .owl). Please email these to me.
2. Some sort of documentation about your classes and definitions. This can be very brief, but if you modified any of the "specificTumorExamples", then you must explain how and why you made changes. If you add additional specificTumorExamples, or additional sets of classes elsewhere in the ontology, brief explanations of these would be nice too.

Grading criteria:

As the name of the owl project implies, what I've given you is an ablated version of a complete OWL ontology that correctly classifies all of the example of specific tumors. Thus, my primary grading criteria is that your ontology also correctly classifies the tumors into their correct stages.

Even if you do not "correctly" classify the examples, I would hope your ontology is consistent. It should be both consistent as assessed by RACER, and consistent with some interpretation of the AJCC description of lung cancer staging. Partial credit will be given to assignments that classify at least some of the examples correctly, so you may want to work on the "simpler" definitions first.