Design Choices and Issues in creating the Bench to Bedside Ontology
1. Should we use relationships or classes to model knowledge?
- Relationship transcribed_to
- Modeling disease as a process or relationship between two physiologic states of a patient?
2. Should more specific pieces of knowledge be modeled as instances or subclasses?
- Parkinson's Disease instance-of/subclass-of Disease
- Alpha Synuclein, Parkin Gene instance-of/subclass-of Gene
- Synuclein, Parkin Protein, DJ-1, PINK-1 instance-of/subclass-of Protein
3. What level of granularity should the knowledge be represented?
- Should the relationship causes/caused-by be inherited to each of the Gene/Disease subclasses and represented at lower levels of granularity
- What is the role of transitivity of relationships, for e.g., a protein is a component of physiologic structure after applying transitive closure.
- Can a relationship such as pathological hallmark be generalized, for e.g., to the level of Disease, Physiologic Structure.
- Can a relationship such as risk factor be generalized to the level of Disease and Genes?
4. Ontology Partitioning and Modularity
- Is it feasible to have an ontology module for Study on one hand and Ontological Elements on the other?
- Would it make sense to have another module for Experimental Data?
- AJ Chen's and Tim Clark's SPE Ontology could be viewed as a separate module within the Bench to Bedside Ontology?
- Can the ACPP ontology be linked to the Bench to Bedside Ontology as a submodule?
- Ontology inclusion: Can we include properties from other ontologies? Is a "shallow copy" enough? Or do we need to copy the axioms that constrain the representation as well? How do we ensure that they are not inconsistent with the rest of the current ontology?
- Ontology subsetting: Can subsets of ontologies be imported into another ontology or do other ontologies have to be imported?
- Based on the Seed Ontology: How do we specify cross links? So if we cross link to something like NN:529, does it pull in the whole subtree
5. Treatment of Properties
- How do we handle multiple domains of a property?
- How do we handle multiple ranges of a property?
- Gene has_variant AllelicVariant violates the ALL/SOME structure, Not all copies of a particular gene have an allelic variant
- Gene transcribes_into Protien violates the ALL/SOME structure, Not all copies of a gene transcribe into a protein
- What is the interpretation of a property: ALL/SOME, ALL/ALL, SOME/SOME, or SOME/ALL?
6. Domain Specific Knowledge Modeling Issues
- Is Protein Degradation a function of the Pathway or the proteosome?
- How does one model protein folding, processing and regulation?
- Can the following constraint be represented in the ontology: "simple overexpression of wild type protein is sufficient to cause Parkinson’s Disease”
6. Role of Implementation Consideration in determining Ontology Creation Best Practices
- Should the decision of whether to model something as a class or instance depend on the performance considerations of OWL reasoners? This can be further clarified as follows:
- What will be the Abox inferences implemented in the context of "instance-of" representation?
- What will be the Tbox inferences implemented in the context of "subclass-of" representation?
- What are the performance and scalability implications of the above choices?
- What are the expressivity implications of the above choices; i.e., can we express some knowledge
using subclass-of based modeling which are not possible using instance-of modeling; or vice versa....
7. OWL Language Issues:
- How does one handle initial/default values, for e.g., function of proteosomal pathway is protein degradation
- How does one represent ternary relationships such as GeneDiseaseAssociationInStudy ?