OBO

http://www.obofoundry.org/

OBO Format

• Models a subset of the semantics available in OWL 2.
• Human-readable flat file format.
• Most current software also supports JSON, but we will stick with OBO for a moment to introduce the basics!
• Begins with a header:

format-version: 1.2
date: 29:07:2010 12:36
auto-generated-by: OBO-Edit 2.0
subsetdef: goslim_candida "Candida GO slim"
subsetdef: goslim_generic "Generic GO slim"
default-namespace: gene_ontology

RDF

• Requires a system of machine-processable identifiers for identifying resources that are unique across the entire Web.
• Requires a computer language for representing and processing these statements.
• The existing Web architecture provides URLs (Uniform Resource Locators) as one form of identifier.
  • Example: For http://www.example.org, the location is www.example.org and the protocol is http.

RDF: URLs, URIs, and URNs

• The other class is the Uniform Resource Name (URN).
• A URN represents a name for a resource that is unique and global throughout the Web.
• Example: urn:isbn:0691141339 (ISBN for a specific book).
• Key Difference: While this URI is unique, it cannot be used to retrieve the book from the Web directly.

RDF: Serialization

• Many RDF files are written in RDF/XML.
• The following example is modified from an RDF file created by the Uniprot consortium.

RDF: Serialization

1. <?xml version='1.0' encoding='UTF-8'?>
2. <rdf:RDF xmlns="[http://purl.uniprot.org/core/](http://purl.uniprot.org/core/)" 
3.          xmlns:rdf="[http://www.w3.org/1999/02/22-rdf-syntax-ns#](http://www.w3.org/1999/02/22-rdf-syntax-ns#)" 
4. >
5. 
6.   <rdf:Description 
7.        rdf:about="[http://purl.uniprot.org/enzyme/1.14.11.2](http://purl.uniprot.org/enzyme/1.14.11.2)">
8.     <rdf:type rdf:resource="[http://purl.uniprot.org/core/Enzyme](http://purl.uniprot.org/core/Enzyme)"/>
9.     <name>Procollagen-proline dioxygenase</name>
10.    <name>Procollagen-proline 4-dioxygenase</name>
11.    <name>Prolyl 4-hydroxylase</name>
12.    <activity>L-proline-[procollagen] + 2-oxoglutarate + O(2)
13.              = trans-4-hydroxy-L-proline-[procollagen] + 
14.                succinate + CO(2).
15.    </activity>
16.    <cofactor>Iron</cofactor>
17.    <cofactor>L-ascorbic acid</cofactor>
18.   </rdf:Description>
19. </rdf:RDF>

• Line 1: The XML declaration,
• Line 2: Opens an element, which indicates that the XML content from line 2 up to the closing tag at line 19 represents RDF.
• Line 2: The default name space for this file is http://purl.uniprot.org/core/
• Line 3: The namespace declaration (xmlns) states that the prefix is equivalent to the namespace URI , and allows the prefix to be used in place of the complete URL in the remainder of the RDF file.
• Lines 6-18: Contain the specific information of the RDF file.

RDF: rdf:about

<?xml version='1.0' encoding='UTF-8'?>
<rdf:RDF xmlns="http://purl.uniprot.org/core/" 
         xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#">
 
  <rdf:Description 
       rdf:about="http://purl.uniprot.org/enzyme/1.14.11.2">
    <rdf:type rdf:resource="http://purl.uniprot.org/core/Enzyme"/>
    <name>Procollagen-proline dioxygenase</name>
    <name>Procollagen-proline 4-dioxygenase</name>
    <name>Prolyl 4-hydroxylase</name>
    <activity>L-proline-[procollagen] + 2-oxoglutarate + O(2)
              = trans-4-hydroxy-L-proline-[procollagen] + 
              succinate + CO(2).
    </activity>
    <cofactor>Iron</cofactor>
    <cofactor>L-ascorbic acid</cofactor>
  </rdf:Description>
</rdf:RDF>

• RDF statements are descriptions of resources
• rdf:about is a fundamental attribute used to identify the Subject of a description.
• Line 6: Indicates that the subject of the RDF statements contained within the Description element is http://purl.uniprot.org/enzyme/1.14.11.2.
• We will abbreviate the subject as :1.14.11.2 in the following slides

RDF: rdf:type

<?xml version='1.0' encoding='UTF-8'?>
<rdf:RDF xmlns="http://purl.uniprot.org/core/" 
         xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#">
 
  <rdf:Description 
       rdf:about="http://purl.uniprot.org/enzyme/1.14.11.2">
    <rdf:type rdf:resource="http://purl.uniprot.org/core/Enzyme"/>
    <name>Procollagen-proline dioxygenase</name>
    <name>Procollagen-proline 4-dioxygenase</name>
    <name>Prolyl 4-hydroxylase</name>
    <activity>L-proline-[procollagen] + 2-oxoglutarate + O(2)
              = trans-4-hydroxy-L-proline-[procollagen] + 
              succinate + CO(2).
    </activity>
    <cofactor>Iron</cofactor>
    <cofactor>L-ascorbic acid</cofactor>
  </rdf:Description>
</rdf:RDF>

• Line 7 contains the first of several triplets contained in the Description element.
  • For each triplet, the subject is 1.14.11.2.
  • The predicate rdf:type means that the subject of the triple is an instance of the class represented by the object of the triple.
• Thus, :1.14.11.2 is an instance of Enzyme

RDF: structural metadata (what something is) vs descriptive data (what something is called or does).

<?xml version='1.0' encoding='UTF-8'?>
<rdf:RDF xmlns="http://purl.uniprot.org/core/" 
         xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#">
 
  <rdf:Description 
       rdf:about="http://purl.uniprot.org/enzyme/1.14.11.2">
    <rdf:type rdf:resource="http://purl.uniprot.org/core/Enzyme"/>
    <name>Procollagen-proline dioxygenase</name>
    <name>Procollagen-proline 4-dioxygenase</name>
    <name>Prolyl 4-hydroxylase</name>
    <activity>L-proline-[procollagen] + 2-oxoglutarate + O(2)
              = trans-4-hydroxy-L-proline-[procollagen] + 
              succinate + CO(2).
    </activity>
    <cofactor>Iron</cofactor>
    <cofactor>L-ascorbic acid</cofactor>
  </rdf:Description>
</rdf:RDF>

• Line 7 contains an rdf:type predicate stating that :1.14.11.2 is an instance of Enzyme
• Line 8 contains a user-defined predicate stating that a name of :1.14.11.2 is Procollagen-proline dioxygenase

RDF: Serialization

RDF: Model

For instance, the triple :1.14.11.2 rdf:type :Enzyme can be represented as shown here:

Figure 1: Graphical Representation of an RDF Triple. The subject, :1.14.11.2, and the object, :Enzyme, are shown as nodes, while the predicate is shown as an edge leading from the subject to the object.

RDF: bnodes

• Suppose we want to represent something that does not have a URI.
• This can be done using an RDF construct called a blank node (bnode for short).
• Let us say we want to express the statement Someone called Douglas wrote a book entitled “Hitchhiker’s Guide to the Galaxy.” Clearly, it is impossible to provide a URI for the concept Someone called Douglas.
• Instead, we use a bnode to express the statement about this “someone.”

[ :firstname "Douglas"]

RDF: bnodes (Continued)

• It is convention to leave a space after the opening square bracket as if to indicate the presence of the “blank” subject of these triples.
• We can now use this statement as an rdf:subject.

[ :firstname "Douglas"] dc:wrote 
         [ dc:title "Hitchhiker's Guide to the Galaxy"] .

RDF: reification

RDF represents a reified statement as four statements using RDF properties and objects: the triplet (S, P, O), reified by resource R, is represented by:

• R rdf:type rdf:Statement
• R rdf:subject S
• R rdf:predicate P
• R rdf:object O

Essentially, RDF “reifies” (makes) a triple into a “thing” about which other RDF statements can be made. The triple is assigned an identifier and treated as a resource.

RDF: reification (Example)

Assuming that the appropriate qname prefixes have been defined, we can express our statement about Wikipedia’s claim as follows:

@prefix rdf: [http://www.w3.org/1999/02/22-rdf-syntax-ns#](http://www.w3.org/1999/02/22-rdf-syntax-ns#) .

_:x rdf:type rdf:Statement .
_:x rdf:subject :"Douglas Adams" .
_:x rdf:predicate :wrote .
_:x rdf:object :"Hitchhiker's Guide to the Galaxy" .

web:wikipedia q:says _:x

RDFS

• RDF Schema (RDFS) is intended as a framework for interpreting the meaning of data expressed in RDF.
• RDFS provides additional specifications and keywords that allow many kinds of inference to be performed.

RDFS: is a ain’t always is a

It is important to distinguish between two different usages of the phrase “is a” in the English language:

• Mickey is a mouse: This means a particular individual called Mickey is an instance of the class Mouse.
• A mouse is a rodent: This means the class “mouse” is a subclass of the class “rodent,” implying that every being that is a mouse is also a rodent.

As we shall see, RDFS has a different syntax for expressing each relation.

RDFS: is a

• It is already possible in RDF to express the statement Mickey is a mouse:

x:Mickey rdf:type x:Mouse .

RDFS: is a (Subclasses)

• The RDFS keywords rdfs:Class and rdfs:subClassOf can be used for this purpose.
• We first declare that x:Mouse and x:Rodent are RDFS classes, and then state that x:Mouse is a subclass of x:Rodent.

@prefix x: [http://www.my-example.org](http://www.my-example.org) .

x:Mouse rdf:type rdfs:Class .
x:Rodent rdf:type rdfs:Class .
x:Mouse rdfs:subClassOf x:Rodent .

RDFS: is a (Inference Rule 1)

One of the RDFS inference rules states that if there are asserted triples of the form:

X rdfs:subClassOf Y .
b rdf:type X .

then the following triple can be inferred:

b rdf:type Y .

• Thus, we can infer that Mickey is an instance of the class !

RDFS: is a (Inference Rule 2)

Another inference rule states that the subclass relation is transitive.

That is, if there are asserted triples about classes \(X\), \(Y\), and \(Z\):

X rdfs:subClassOf Y .
Y rdfs:subClassOf Z .

then the following triple can be inferred:

X rdfs:subClassOf Z .

• Thus, we can infer that mouse is a subclass of the class rodent!

RDFS: is a (Transitivity Example)

• if we assert the following triples:

x:Mouse rdf:type rdfs:Class .
x:Rodent rdf:type rdfs:Class .
x:Mammal rdf:type rdfs:Class .

x:Mouse rdfs:subClassOf x:Rodent .
x:Rodent rdfs:subClassOf x:Mammal .

then we can infer that mice are mammals:

x:Mouse rdfs:subClassOf x:Mammal .

RDFS: Properties and Inference

Similar inference rules apply to properties. For instance, we can define a simple hierarchy of biochemical regulation with the following RDF statements:

@prefix r: [http://www.regulation.org](http://www.regulation.org) .

r:regulates rdf:type rdfs:Property .
r:positively_regulates rdf:type rdfs:Property .
r:negatively_regulates rdf:type rdfs:Property .

r:positively_regulates rdfs:subPropertyOf r:regulates .
r:negatively_regulates rdfs:subPropertyOf r:regulates .

If proteins A and B have been defined in a namespace pro, we can state that A positively regulates B by asserting the following triple:

pro:A  r:positively_regulates pro:B .

RDFS: Properties and Inference

There is an inference rule for RDFS properties stating that if the following triples are asserted:

P rdfs:subPropertyOf Q .
X P Y .

Then the following triple can be inferred:

X Q Y .

Example: Using our regulation hierarchy, we can infer that Protein A regulates Protein B:

pro:A  r:regulates pro:B .

RDFS: Property Domains

• RDFS’ use of domain and range enables inference.
• The statement P rdfs:domain D means that the subject of any triple using predicate P must be an instance of class D.

If the following two triples are asserted:

P rdfs:domain D .
X P Y .

hen the following triple is inferred:

X rdf:type D .

RDFS: Property Ranges

Similarly, the RDFS statement P rdfs:range R for an rdf:Property P and an rdf:Class R means that the resource denoted by the object of a triple whose predicate is P is an instance of the class R.

If the following two triples are asserted:

P rdfs:range R .
X P Y .

Then the following triple can be inferred:

Y rdf:type R .

RDFS: Property Domains and Ranges (Example)

:drug_affects_protein rdf:type rdfs:Property .
:Drug rdf:type rdfs:Class .
:Protein rdf:type rdfs:Class .

:drug_affects_protein rdfs:domain :Drug .
:drug_affects_protein rdfs:range :Protein .

If we now additionally assert the following triple:

:Nilotinib  :drug_affects_protein :BCR-ABL .

Then we can infer the following two triples:

:Nilotinib rdf:type :Drug .
:BCR-ABL rdf:type :Protein .

RDFS: Property Domains and Ranges (Graph)

The figure below displays the graph corresponding to these triples. The inferred triples are shown using dashed lines.

Figure 3: Example of RDFS Inference in an RDF Graph. Because of the domain and range restrictions on the predicate :drug_affects_protein, it can be inferred that :Nilotinib is a drug and :BCR-ABL is a protein.

RDFS: Intersections and Unions

• If \(Z\) is a subclass of both class \(X\) and class \(Y\), and \(b\) is defined as an instance of \(Z\), then \(b\) can be inferred to be an instance of both \(X\) and \(Y\).

If the following triples are asserted:

Z rdfs:subClassOf X .
Z rdfs:subClassOf Y .
b rdf:type Z .

Then these two triples can be inferred:

b rdf:type X .
b rdf:type Y .

RDFS: Intersections and Unions (Continued)

For instance, if we state that:

:Pig rdfs:subClassOf :Mammal .
:Pig rdfs:subClassOf :FarmAnimal .

Then we can conclude that a particular pig is both a mammal and a farm animal. The class :Pig can be considered to occupy the intersection between those two classes.On the other hand, asserting that:

Cow rdfs:subClassOf :Mammal .
:Cow rdfs:subClassOf :FarmAnimal .

Does not imply that a cow is a pig!

RDFS: Intersections and Unions (Union)

If we assert the following triples:

X rdfs:subClassOf Z .
Y rdfs:subClassOf Z .

Then, if either b rdf:type X . or b rdf:type Y . is asserted, it can be inferred that:

b rdf:type Z .

The Web Ontology Language (OWL)

The acronym OWL stands for Web Ontology Language. While the most natural acronym would have been “WOL,” OWL was chosen because it is easier to pronounce.¹

In essence, OWL is an ontology language that builds upon RDF and RDFS but provides more powerful inference rules and built-in constructs.

• OWL can be understood as an extension of RDFS that allows more powerful inference to be performed.
• To a first approximation, we can say that OWL extends the inference capabilities already present in RDFS by the addition of new constructs and new rules.

Defining Classes in OWL

OWL classes can be defined in two primary ways:

• Enumerated Collections: Classes representing a specific, closed collection of individuals (e.g., Paul, John, and Mary).
• Conceptual Definitions: Classes representing concepts that can have an indeterminate number of instances (e.g., Mitochondrion).

Defining Classes in OWL

OWL provides six different mechanisms for defining classes:

1. Named Classes: Indicated by a URI.
2. Enumerations: A list of all specific individuals in the class.
3. Property Restrictions: Defining a class by how its members relate to other things.
4. Intersection: The overlap of two or more class descriptions.
5. Union: The combination of two or more class descriptions.
6. Complement: Everything that does not belong to a specific class.

Aside from the first type, each of these mechanisms defines classes by placing restrictions on the things that belong to the class.

Classes in OWL: Enumeration

For instance, the following class definition expressed using RDF/XML syntax creates a class containing exactly the four Beatles:

<owl:Class>
  <owl:oneOf rdf:parseType="Collection">
    <owl:Thing rdf:about="#John"/>
    <owl:Thing rdf:about="#Paul"/>
    <owl:Thing rdf:about="#George"/>
    <owl:Thing rdf:about="#Ringo"/>
  </owl:oneOf>
</owl:Class>

Classes in OWL: Property Restriction

<owl:Class rdf:about="#A-Microtubule">
  <rdfs:label>A-Microtubule</rdfs:label>
  <rdfs:subClassOf rdf:resource="#Cilium Microtubule"/>
  
  <rdfs:subClassOf>
    <owl:Restriction>
      <owl:onProperty rdf:resource="#is Physical Part of"/>
      <owl:someValuesFrom rdf:resource="#Cytoskeleton"/>
    </owl:Restriction>
  </rdfs:subClassOf>

  <rdfs:subClassOf>
    <owl:Restriction>
      <owl:onProperty rdf:resource="#is Physical Part of"/>
      <owl:someValuesFrom rdf:resource="#Cilium"/>
    </owl:Restriction>
  </rdfs:subClassOf>
</owl:Class>

Classes in OWL: Property Restriction (Logic)

• The previous RDF/XML code states that the class A-Microtubule is a subclass of an anonymous class defined by being a physical part of some cytoskeleton.
• It is also a subclass of another anonymous class defined by being a physical part of some cilium.

In plain English: Every A-microtubule is part of a cilium and simultaneously part of a cytoskeleton.

Classes in OWL: Property Restriction (Value Constraints)

OWL distinguishes two kinds of property restrictions: Value Constraints and Cardinality Constraints. A value restriction on some property \(P\) has the following general form:

<rdfs:subClassOf>
  <owl:Restriction>
    <owl:onProperty rdf:resource="P" />
    <owl:allValuesFrom rdf:resource="#V"/>
  </owl:Restriction>
</rdfs:subClassOf>

Classes in OWL: Property Restriction (Universal)

• Confusing Caveat: owl:allValuesFrom does not necessarily mean that an instance of the class must have the property \(P\). It merely states that if an instance does have \(P\), then the value must be from class \(V\).
• The Universal Quantifier: This is analogous to the logic of \(\forall\) in predicate calculus.
• Formal Logic: Recall that the formula \((\forall x) P(x)\) essentially means “for all \(x\), if \(x\) exists in this context, then it must satisfy \(P\).”

Classes in OWL: Negation

• For instance, the Ontology for Biomedical Investigations (OBI) describes a class of cells that do not express the CD8 receptor using the owl:complementOf keyword.
• This is achieved by defining the intersection of the class “Cell” and the complement of the class of things that “have part CD8 receptor.”

<owl:intersectionOf rdf:parseType="Collection">
  <owl:Class rdf:about="CL:cell"/>
  
  <owl:Class>
    <owl:complementOf>
      <owl:Restriction>
        <owl:onProperty>
          <owl:TransitiveProperty rdf:about="#has_part"/>
        </owl:onProperty>
        <owl:someValuesFrom>
          <owl:Class rdf:about="#CD8 receptor"/>
        </owl:someValuesFrom>
      </owl:Restriction>
    </owl:complementOf>
  </owl:Class>
</owl:intersectionOf>

Classes in OWL: Negation (The Pitfalls)

• The Logic: While one might be tempted to think that the complement of Man is Woman, the formal meaning of owl:complementOf Man is everything in the universe that is not a Man.
  • This includes the kitchen sink, the planet Neptune, and the Pacific Ocean.
• The Solution: owl:complementOf is almost always combined with other restrictions (intersections) to narrow the scope.

Example: To define Woman, you would use the Intersection of Human Being and the Complement of Man.

Classes in OWL: Union

• For instance, we can define the class Human as the union of the classes Man and Woman:

<owl:Class rdf:ID="Human">
  <owl:unionOf rdf:parseType="Collection">
    <owl:Class rdf:about="#Man" />
    <owl:Class rdf:about="#Woman" />
  </owl:unionOf>
</owl:Class>

Classes in OWL: Union (The Subclass Trap)

Contrast the owl:unionOf construct to the following incorrect attempt to define Human as a subclass of Man and Woman:

<owl:Class rdf:ID="Human">
  <rdfs:subClassOf rdf:resource="#Man" />
  <rdfs:subClassOf rdf:resource="#Woman" />
</owl:Class>

Why this fails:

• The “And” vs. “Or” Logic: By using rdfs:subClassOf twice, you are stating that every Human must be a Man AND simultaneously a Woman.
• Unsatisfiability: If Man and Woman are defined as disjoint (meaning they cannot overlap), this class becomes unsatisfiable.
• The Empty Set: An unsatisfiable class is one that cannot contain any possible individuals.

In formal logic, this is equivalent to the empty set (\(\emptyset\)).

Individuals in OWL

In OWL, rdf:type is the property that connects an individual to a class of which it is a member.

If we have a class of cities:

<owl:Class rdf:ID="City"/>

We can declare individual cities to be members of this class. The following two syntactical forms have identical meanings:

• Option 1: Direct Class Tag

This uses the class name as the XML element tag.

<City rdf:ID="Berlin"/>

• Option 2: Explicit Typing

This uses owl:Thing (the root class of everything) and explicitly states the type.

<owl:Thing rdf:ID="Berlin"/>

<owl:Thing rdf:about="#Berlin">
  <rdf:type rdf:resource="#City"/>
</owl:Thing>

OWL: Conclusion and Next Steps

• The Tip of the Iceberg: We have barely scratched the surface of OWL’s logical capabilities in this lecture.
• Hands-on Learning: You will create your own ontology in the upcoming Exercise, where you will have the chance to explore these constructs (and more) in a practical setting.
• Official Resources: The W3C Website provides comprehensive documentation and technical specifications for the OWL family of languages.

Semantic Web Layer Cake