Defined versus Asserted Classes: Working with the OWL Ontologies NIF Webinar February 9 th 2010
Outline NIFSTD ontologies in brief Multiple vs Single hierarchy of classes/ Asserted vs Inferred classes/primitive and Defined classes Simple inference example NIF s Neuron by neurotransmitter classification NIF s Neuron by Brain region classification Bridge files and modularity Searching Neurons through NIF s GWT search interface
NIFSTD Modules Fig.1: The semantic domains (in oval) covered in the NIFSTD with some of the subdomains (in rectangle). Each of the domains are covered by a separate OWL module Overview. Constructed based on the best practices closely followed by the Open Biomedical Ontologies (OBO) community Built in a modular fashion, covering orthogonal neuroscience domain e.g. anatomy, cell types, techniques etc. promotes easy extendibility Avoids duplication of efforts by conforming to standards that promote reuse Modules are standardized to the same upper level ontologies The Basic Formal Ontology (BFO), OBO Relations Ontology (OBO-RO), and the Ontology of Phenotypical Qualities (PATO)
Ontology Adopted to CS by AI community as explicit specification of conceptualization (T. Gruber) Organizing the concepts involved in a domain to a hierarchy and Precisely specifying how the concepts are interrelated with each other Explicit knowledge are asserted but implicit consequences should rely on reasoners
OWL-DL NIFTSD ontologies are represented in OWL-DL language Standard language defined by (W3C) Largely influenced by Description Logics Decidable fragment of First Order Logic Useful reasoning services from common reasoner such as Pallet, Racer Pro, Fact++ etc. Automatic Subsumption/ Classification Consistency checking Using a reasoner to classify the class hierarchy is a powerful feature of building an ontology using the OWL-DL
Asserted vs. Inferred classes NIFSTD chose single inheritance principle Class hierarchies are constructed as a simple tree Asserted hierarchy (manually created hierarchy) should have only one super class. It keeps the classes univocal and avoids ambiguity By asserted hierarchy we would mean a hierarchy that represents a universal facts in the BFO sense OBO foundry recommendation We are aware that there are cases where multiple parents are required. Example: the universal fact about Purkinje cell can be that it is a kind of Neuron. However, the same cell can have more specific views such as it s a GABAergic neuron or it s kind of a Cerebellum neuron. Single inheritance is often misunderstood to mean that you can only have a single parent Multiple parents can actually be derived/ inferred in a logical way Rely on automated reasoning to compute and maintain multiple inheritence
Asserted vs. Inferred classes Reasoners can keep the hierarchies in a maintainable and logically correct state Provides a logical and intuitive reason as to how a class X may exist in multiple/different hierarchies Saves a great deal of manual labor Minimizes human errors as well Keeps the ontology in a maintainable and modular state Promotes the reuse of the ontology by other ontologies and applications
Primitive and Defined Classes Primitive classes Has a set of necessary conditions Defined classes Has a set of necessary and sufficient restrictions; defined by equivalent statement in OWL. Automated classification is possible on defined classes through reasoners
DL Reasoning Example Defined Classes Woman Person hasgender. Female Mother Parent haschild. Person hasgenderfemale. Parent Person haschild. Person Relations/ Properties: haschild (Person, Person) hasgender (Person, Gender) Parent Person haschild. Person [ FOL : Parent ( x) Person ( x) y( haschild ( x, y) Person ( y))] 9
DL Reasoning Example 10
NIF s Neuron Classifications List of NIF neurons in NeuroLex (wiki version of NIFSTD) http://neurolex.org/wiki/category:neuron We wanted to classify the neurons based on their Neurotransmitter and also based on their soma location in different brain regions Neuron by Neurotransmitter http://neurolex.org/wiki/neuron_by_neurotransmitter Neuron by region http://neurolex.org/wiki/neuron_by_region
Bridge files NIFSTD NIF-Cell NIF- Subcellular NIF- Anatomy NIF- Molecule NIF-Neuron-BrainRegion-Bridge.owl NIF-Neuron-NT-Bridge.owl Cross-module relations among classes are assigned in a separate bridging module. Allows different users to assert their own restrictions in a different bridge file without worrying about NIF-specific view of the restriction on core modules.
Neuron by Neurotransmitter Classification Based on NeuroLex wiki contributions by NIF cell working group, a bridge file has been constructed between NIF-Cell and NIF- Molecule Assigned relation between a neuron and its neurotransmitter Defined classes to generate an inferred classifications of Neurons by their neurotransmitters (e.g., GABAergic neurons, Glutamatergic neurons etc.) Currently using a macro relation called has_neurotransmitter. This relation will be further defined in terms of other obo relations to associate other intermediate concepts Ex: x has_neurotransmitter y <=> x has_disposition some (realized_as some (GO:synaptic_transmission and has_participant some (y and has_role neurotransmitter_role))); [As proposed by Chris Mungall] Bridge file location: http://ontology.neuinfo.org/nif/biomaterialentities/nif-neuron-nt- Bridge.owl
Neuron by Brain Region Classification We ve created another bridge file based on NeuroLex contributions Assigns relations between a neuron and its soma location in different brain regions Defined Neurons based on their brain region, e.g., Hippocampal neuron, Cerebellum neuron, Neocortical neuron etc. We have a macro relation has_soma_location and corresponding actual relation: x has_soma_location y <=> neuron_type_x has_part some ('somatic portion' and (part_of some brain_region_y)); Location of the Bridge file: http://ontology.neuinfo.org/nif/biomaterialentities/nif- Neuron-BrainRegion-Bridge.owl
Example Neurons with Necessary Restrictions
Defined Neuron Classes Example
Demos in Protégé
Neurons through NIF GWT http://nif-apps-stage.neuinfo.org/nif/nifgwt.html
Acknowledgement NIF-Cell working group: Giorgio Ascoli, Gordon Shepherd, Sridevi Polavar, Stephen Larson, MaryAnn Martone