Transcript Slide 1

Ontologies:
“What are they?” and “How do they work?”
Michael Grobe
(work supported in part by
Research Technologies
UITS
Indiana University)
1
Table of Contents
Panorama of definitions
Explication of the Big Definition
The Gene Ontology as an example
Processing queries on data annotated with
ontology classifications
Merging and building ontologies
Table of Non-contents
Automated annotation
The role of ontologies in the Semantic Web
Using ontologies in bioinformatics research
2
Panorama of definitions of “Ontology”
In standard use: “Ontology is “is a study of conceptions
of reality and the nature of being. … It is the science of
what is, of the kinds and structures of the objects,
properties and relations in every area of reality.”
(Wikipedia, 2008)
The term was hijacked for use within information
science (sic) where it has many applications, but…
“People use the word ontology to mean different things,
e.g.
- glossaries and data dictionaries,
- thesauri and taxonomies,
- schema and data models, and
- formal ontologies and inference.” (Pidcock, 2003)
3
More definitions
Here’s a definition from Uschold, et al. quoted by
Stevens, et al.:
“An ontology may take a variety of forms, but necessarily
it will include a vocabulary of terms, and some
specification of their meaning.
This includes definitions and an indication of how
concepts are inter-related which collectively impose a
structure on the domain and constrain the possible
interpretations of terms.”
4
More definitions
And a definition from Pidcock, 2003:
“A formal ontology is a controlled vocabulary expressed
in an ontology representation language.
This language has a grammar for using vocabulary
terms to express something meaningful within a
specified domain of interest.
The grammar contains formal constraints … on how
terms in the ontology’s vocabulary can be used
together.”
5
A definition and a “clarification”
And another definition by Gruber, 1993:
“I use the term ontology to mean the specification of a
conceptualization. …A conceptualization is an
abstract, simplified view of the world that we wish to
represent for some purpose.”
Stevens, et al., 2000 clarify (?):
“The conceptualisation is the couching of knowledge
about the world in terms of entities (things, the
relationships they hold and the constraints between
them). The specification is the representation of this
conceptualisation in a concrete form.”
6
Yet more definitions
And Gruber defines one purpose for ontologies:
“Ontologies provide controlled, consistent vocabularies
to describe concepts and relationships, thereby
enabling knowledge sharing” (Gruber, 1993)
But then, one also finds descriptions like:
“Shallow ontologies comprise relatively few unchanging
terms that organize very large amounts of data—for
example, terms such as customer, account number, and
overdraft…” (Shadbolt, 2006)
7
And Wikipedia continues in this vein by
describing 2 types of (IT-related) ontology:
“A domain ontology (or domain-specific ontology)
models a specific domain, or part of the world. It
represents the particular meanings of terms as they
apply to that domain.” (Wikipedia, 2008)
“An upper ontology (or foundation ontology) is a model
of the common objects that are generally applicable
across a wide range of domain ontologies. It contains a
core glossary in whose terms objects in a set of
domains can be described.” There are several
standardized upper ontologies available for use,
including Dublin Core . . . (Wikipedia, 2008)
8
As an aside (because we are actually interested
in domain ontologies), let’s take a quick look at
the Dublin Core thanks again to Wikipedia:
“The Simple Dublin Core Metadata Element Set
(DCMES) consists of 15 metadata elements:
Title
Description
Date
Identifier
Relation
Creator
Publisher
Type
Source
Coverage
Subject
Contributor
Format
Language
Rights“
Surprisingly, these are described as “metadata
elements”.
9
But wait, there’s still more…
In particular, an ontology, or some ontologies, provide
some ability to “reason”:
“…an ontology is a representation of a set of concepts
within a domain and the relationships between those
concepts. It is used to reason about the properties of
that domain, and may be used to define the domain.”
(Wikipedia, 2008)
“. . . a document or file that formally defines the
relations among terms. The most typical kind of
ontology for the Web has a taxonomy and a set of
inference rules.” TimBL, 2000.
10
Note that this “reasoning” is performed using
terms representing concepts rather than the
concepts themselves. (Which is to say, text
strings are being shuffled around; there is no
“thought” involved.)
“The computer doesn’t truly “understand” any of
this information, but it can now manipulate the
terms much more effectively in ways that are
meaningful to the human users.” (TimBL, 2001)
11
Note also that this is the first definition that refers to a “taxonomy”,
and the term comes up a lot in the field, so let’s look at an example
taxonomy.
Consider (a simplified version of) the Linnaean classification of
living organisms based on these categories:
Dominion
Domain
Kingdom
Phylum
Subphylum
Class
Cohort
Order
Family
Genus
Species
Each individual organism is assigned a set of values, one for each
category. The result is a large table with 11 (or more) columns.
This taxonomy defines a hierarchy of sets and subsets, and . . .
. . . the series of values in each column of an individual species
record represents the path from the root of the tree to that species
(leaf node), and much of this path information is redundant.
12
There may be better ways to store this redundant data,
and . . .
. . . there may be other “ways” to think about what the
data mean.
In particular, the set-subset relationships may be
thought of as “inference rules” that can be applied to
answer queries. For example:
If
an organism is a member of a genus
and
that genus is a member of a family
then
that organism is also a member of that family
13
Finally, here is a “big,” formal definition:
"An ontology O is a six-tuple C, HC, HR, L, FC, FR,
where
C is the set of concepts,
HC a taxonomy induced on the concepts,
HR the set of non-taxonomic relations,
L the set of terms (lexicals) which refer to concepts
and relations, and
FC, FR are relations that map the terms in L to the
corresponding concepts and relations.
If the ontology is dynamic all these structures are likely
to change over time." (Niepert, et al., 2008)
Note that this definition includes no mention of
“inference,” but inference may be hidden within.
14
This is clearly a complicated description, but we can
break it into parts, at least some of which are
understandable:
First, a couple of definitions:
A “tuple” is a set of objects in a specified order.
An “N-tuple” is a tuple that contains exactly N items.
A “relation” is a set of “tuples” of the same “arity”, but may
be thought of as a “table”, which is how Relational
Databases came to be named (even tho there are differences).
Now note that the primary set, C, is a “set of concepts,” not a “set
of terms”.
“Terms,” are used to “refer to” the “concepts”, and both terms and
relations are likely to change over time.
FC maps L to C, but “terms” also refer to “relations”?
15
As a very simple example, here’s a set of concepts (C)
represented as strings of English text:
{
“Vehicle”, “Car”, “Truck”, “2-wheel drive car”,
“4-wheel drive car”, “front-wheel drive car”,
“rear-wheel drive car”
}
Here’s a “taxonomy” (known as HC, perhaps called “is_a”?)
“induced” on the set of concepts:
{
( “Car”, “Vehicle” ),
( “Truck”, “Vehicle” ),
( “2-wheel drive car”, “Car” ),
( “4-wheel drive car”, “Car” ),
( “front-wheel drive car”, “2-wheel drive car” ),
( “rear-wheel drive car”, “2-wheel drive car” )
}
16
Here’s a set of terms, L = ( 0, 1, 2, 3, 4, 5, and 6 ) and a relation
(FC) mapping terms from the term set to concepts:
{
( 0, “Vehicle” ),
( 1, “Car” ),
( 2, “Truck” ),
( 3, “2-wheel drive car” ),
( 4, “4-wheel drive car” ),
( 5, “4-wheel drive car” ),
( 6, “4-wheel drive car” )
}
Note that it is a good idea to use “meaningless” terms: Identifiers
like “GO:000056”.
Here’s a representation of the taxonomy (HC) using terms:
{ ( 1, 0 ), ( 2, 0 ), ( 3, 1 ), ( 4, 1 ), ( 5, 3 ), ( 6, 3 ) }
17
Here’s a relation (call it “is_transitively_a” or
“is_a_descendent_of” or a “transitive closure”) derived
from the taxonomy assuming “transitivity”:
{
( 1, 0 ),
( 2, 0 ),
( 3, 1 ), ( 3, 0 ),
( 4, 1 ), ( 4, 0 ),
( 5, 3 ), ( 5, 1 ), ( 5, 0 ),
( 5, 3 ), ( 5, 1 ), ( 5, 0 ),
}
The items in bold were added “by transitivity”.
This seems to be one way of “sneaking” inference into
the definition.
18
This complete table, ….ur….relation, contains the same
information as could be inferred from the transitivity
inference rule:
If
Item_A is_a Item_B
and
Item_B is_a Item_C
then
Item_A is_a Item_C
….and in some cases the relation derived by transitivity
would be prohibitively large, so inference rules are
frequently used to determine the relationship between 2
items ad hoc.
Another way to “sneak” inference into this definition
would be to consider the taxonomy as a set of inference
rules, as will be considered later.
19
The Gene “Ontology”
One of the best known “ontologies” is the Gene Ontology which is
actually 3 separate “ontologies” (with different “namespaces”)
- molecular function (cell biochemistry?)
What biochemical reactions do gene products perform?
- biological process (cell physiology?)
What cellular processes do the gene products
participate in?
- cellular component (cell anatomy?)
In which cellular compartments or locations are those
gene products expressed?
20
Here is a portion of the GO is_a DAG (Blake, 2004) for molecular
function (example: “chromatin binding” is_a “DNA binding”):
(It is easy to confuse a gene product name with its molecular function, and for that reason many
GO molecular functions are appended with the word "activity". www.geneontology.org, 2008)
21
Here is a subset (C) of the Gene Ontology molecular
function concepts
binding
enzyme activity
helicase activity
DNA binding
nucleic acid binding
chromatin binding
lamin/chromatin binding
DNA helicase activity
ATP-dependent helicase activity
adenosine triphosphatase activity
ATP-dependent DNA helicase activity
DNA-dependent adenosine triphospatase activity
(It is easy to confuse a gene product name with its molecular function, and for that
reason many GO molecular functions are appended with the word "activity".
www.geneontology.org, 2008)
22
The set (L) of Gene Ontology molecular function
terms
GO:00005488
GO:00008047
GO:00004386
GO:00003677
GO:00003676
GO:00003682
GO:00003683
GO:00003679
GO:00008026
GO:00016887
GO:00004003
GO:00008094
23
The relation (FC) mapping GO terms to concepts
GO:00005488
GO:00008047
GO:00004386
GO:00003677
GO:00003676
GO:00003682
GO:00003683
GO:00003679
GO:00008026
GO:00016887
GO:00004003
GO:00008094
binding
enzyme activity
helicase activity
DNA binding
nucleic acid binding
chromatin binding
lamin/chromatin binding
DNA helicase activity
ATP-dependent helicase activity
adenosine triphosphatase activity
ATP-dependent DNA helicase activity
DNA-dependent adenosine
triphosphatase activity
24
Here is the is_a relation (HC) defining relationships among concepts
(nucleic_acid binding activity “is a kind of” binding activity):
Sub-function
Function
molecular_function
root
binding
molecular function
nucleic acid binding
binding
enzyme activity
molecular function
helicase activity
enzyme activity
DNA binding
nucleic acid binding
chromatin binding
DNA binding
lamin/chromatin binding
chromatin binding
DNA helicase activity
DNA binding
DNA helicase activity
helicase activity
ATP-dependent helicase activity
helicase activity
adenosine triphosphatase activity
enzyme activity
ATP-dependent helicase activity
adenosine triphosphatase activity
DNA-dependent adenosine triphosphatase activity
adenosine triphosphatase activity
ATP-dependent DNA helicase activity
DNA helicase activity
ATP-dependent DNA helicase activity
ATP-dependent helicase activity
ATP-dependent DNA helicase activity
DNA-dependent adenosine
triphosphatase activity
(Note that some Sub-functions have multiple parent Functions.)
25
Here is a portion of the GO is_a DAG (Blake, 2004) for molecular
function (example: “chromatin binding” is_a “DNA binding”):
(It is easy to confuse a gene product name with its molecular function, and for that reason many
GO molecular functions are appended with the word "activity". www.geneontology.org, 2008)
26
Here’s the first entry (of the ~26K) in the GO
text version (with all three parts intermixed):
[Term]
id: GO:0000001
name: mitochondrion inheritance
namespace: biological_process
def: "The distribution of mitochondria, including
the mitochondrial genome, into daughter cells
after mitosis or meiosis, mediated by
interactions between mitochondria and the
cytoskeleton." [GOC:mcc, PMID:10873824,
PMID:11389764]
synonym: "mitochondrial inheritance" EXACT []
is_a: GO:0048308 ! organelle inheritance
is_a: GO:0048311 ! mitochondrion distribution
You can also get the GO as RDF XML, or as a
MySQL database.
27
In the example, a GO concept (“name”) is being mapped to:
- a GO ID,
- a root “namespace”,
- a “def,” and also to a
- set of “synonyms”.
And, in addition, the concept may be mapped to “parent” or “child”
concepts through the
- “is_a” (subsumption) and/or
- “part_of” (meronomy/partonomy),
- “regulates” (gene transcription).
- “positively_regulates”,
- “negatively_regulates” links
as exemplified in the next slide.
Remember that “is_a” is really more like ”is a kind of”, and the last
3 in the list above are “non-taxonomic relations” (HR).
These links define edges of the GO DAGs.
28
[Term]
id: GO:0003677
name: DNA binding
namespace: molecular_function
def: "Interacting selectively with DNA (deoxyribonucleic acid)."
[GOC:jl]
subset: goslim_candida subset: goslim_generic subset: goslim_plant
subset: goslim_yeast
subset: gosubset_prok
related_synonym: "microtubule/chromatin interaction" []
narrow_synonym: "plasmid binding" []
is_a: GO:0003676 ! nucleic acid binding
[Term]
id: GO:0003682
name: chromatin binding
namespace: molecular_function
def: "Interacting selectively with chromatin, the network of fibers of
DNA and protein that make up the chromosomes of the eukaryotic nucleus
during interphase." [GOC:jl, ISBN:0198506732 "Oxford Dictionary of
Biochemistry and Molecular Biology"]
subset: goslim_generic
subset: goslim_pir
subset: goslim_plant related_
synonym: "microtubule/chromatin interaction" []
narrow_synonym: "nuclear membrane vesicle binding to chromatin" []
broad_synonym: "lamin/chromatin binding" []
is_a: GO:0003677 ! DNA binding
(This was changed since 2004.)
29
Note that the Genes listed in the previous DAG
graphic are NOT part of the ontology.
In fact, there is NO “DATA” in the ontology.
Blake (2004) emphasizes some important
features of GO as:
“Not a way to unify biological database[s]
Not a dictated standard
Not a database of gene products, protein domains, or
motifs
Does not define evolutionary relationships”
30
In fact, GO may not even BE an ontology :
“GO ontology, which is more a nomenclature and a
taxonomy, than a formal ontology, is highly successful
and widely used.” (Sheth, 2003)
In fact, that wide usage may be due directly to the fact
that it is NOT a formal ontology:
“Semi-formal ontologies that may be based on limited
expressive power are most practical and useful. Formal
or semi-formal ontologies represented in very
expressive languages…have, in practice, yielded little
value in real world applications.” (Sheth, 2003)
“Our object in touting the value of semi-formal
ontologies is to prevent research in the Semantic Web
field from leading straight into the very problems that AI
found itself in.” (Sheth, 2003)
31
So where is the data?
Here is a 2-column table that uses GO to “annotate” the
products of the genes shown in the graphic above from
the Mouse Genome Initiative database:
Gene Name
Mcmd2
Mcmd4
Mcmd6
Mcmd7
Molecular Function
GO:0003682
GO:0003682,GO:0004003
GO:0003682,GO:0004003
GO:0003682,GO:0004003
Note that only the lowest level GO ID terms are used
here to identify functions.
Note also that a gene product may perform multiple
functions and that multiple function entries in this
table are separated by commas.
32
Scale of the genome annotation
As of August of 2004, the Mouse Genome had been
annotated using the Gene Ontology as:
- Function: 12K genes with 30K annotations
- Process:
11K genes annotated with 21K annotations
- Location:
11K genes annotated with 20K annotations
33
Data may be presented via a tree representation:
binding
(Click an entry to see data
nucleic acid binding
annotated with that entry.)
DNA binding
chromatin binding
lamin/chromatin binding
DNA helicase activity
ATP-dependent DNA-helicase activity
enzyme activity
helicase activity
DNA helicase activity
ATP-dependent DNA helicase activity
ATP-dependent helicase activity
adenosine triphosphatase activity
ATP-dependent helicase activity
ATP-dependent DNA helicase activity
DNA-dependent adenosine triphosphatase activity
ATP-dependent DNA helicase activity
34
When data is annotated using the most specific GO
category, membership in parent categories (supersets)
must to be determined by “inference”, that is, by moving
up the is_a DAG or up the part_of path (if available),
applying the transitivity rule.
We “know” that transitivity holds, and we use it
“intuitively” as we inspect the DAG:
If
X is_a molecular_function_1
and
molecular_function_1 is_a
molecular_function_2
then
X is_a molecular_function_2
35
Or we might think of each entry in the DAG relation as being an
inference rule itself, and apply these rules whenever possible.
So an entry in the is_a DAG like:
( Function_1, Function_2 )
might be interpreted as the inference rule:
If
gene_product_X is annotated with Function_1
and
Function_1 is_a Function_2
then
gene_product_X could be annotated with Function_2
(Aside: In some cases the is_a relation could be interpreted in
reverse order as an “includes” relation?)
36
One way or another:
If
you have the function, process, and location GO IDs for
a collection of genes (which will never be in the GO itself)
and
you have the GO,
and
you have an appropriate inference capability
then
you should be able answer questions that relate to the
membership of any annotated item in any GO
class.
37
How might this process actually work with
questions like:
“Tell me whether mouse Mcmd4 is a helicase.”
which should be roughly equivalent to:
“Is Mcmd4 annotated with “helicase activity” (GO:0004386) or
some child thereof?”
Answer: Yes
“Which mouse genes are involved in DNA binding, but are not DNA
helicases.”
which should be roughly equivalent to:
“Which mouse genes are annotated with “DNA binding”
(GO:0003677) or some child thereof, but are not annotated
with “helicase activity” (GO:0004386) or some child thereof.”
Answer: Mcmd, Mcmd2?
38
We could answer these questions by “inspection”
because we know what the is_a relation “means”, and
how to manipulate the relation “meaningfully”.
However, how can we answer these questions
“mechanically” using a program?
In particular, if we interpret the is_a relation entries as
inference rules, how can we process these queries?
First we will think of the queries as “assertions”, like:
“Mcmd4 is a helicase.”
and
“Gene_product_X displays “helicase activity”
and try to prove (or “satisfy”) them by using the
inference rules to derive a list of facts provable from the
given data.
39
Suppose you want to answer the question:
“Does mouse Mcmd4 display helicase activity?
Start with a “fact base” composed of the set of known “facts” from
your annotation database:
{
( Mcmd4, chromatin binding ),
( Mcmd4, ATP-dependent DNA helicase activity )
}
Then repeatedly apply the inference rules to add facts to the
collection of facts in the “fact base”, and . . .
Stop when the target assertion appears in the fact base, or when
no new facts have been added during a step.
At that point, if the assertion is in the fact base, it is has been
“proved” to be true, else it is false.
(This is a “forward-chaining” inference process.)
40
After step one, the “fact base” will contain
(assuming the entries in the DAG relation are
processed in the order presented earlier):
{
( Mcmd4, chromatin binding ),
( Mcmd4, ATP-dependent DNA helicase activity )
( Mcmd4, DNA helicase activity ),
( Mcmd4, DNA binding activity )
( Mcmd4, ATP-dependent helicase activity ),
( Mcmd4, DNA-dependent adenosine
triphosphate activity )
}
41
After step two, the fact base will contain:
{
( Mcmd4, chromatin binding ),
( Mcmd4, ATP-dependent DNA helicase activity )
( Mcmd4, DNA helicase activity ),
( Mcmd4, DNA binding activity ),
( Mcmd4, ATP-dependent helicase activity ),
( Mcmd4, DNA-dependent adenosine triphosphate activity ),
( Mcmd4, helicase activity ),
( Mcmd4, adenosine triphosphatase activity ),
( Mcdm4, nucleic acid binding )
}
at which point we can stop, because the assertion has been “proved”.
To resolve the second query we initialize the fact base with the entire
annotation database, infer new facts until no facts can be added and
then list the facts that include gene-products with “helicase activity”.
(Aside: How would this be done using SQL?)
42
How would this be done using SQL?
It might be possible to use a series of self-joins to get
records that include the full path from each concept to
root.
On the other hand, it would probably be better to compute
the paths using some external tool and store the result as
a table of concept-ancestor pairs (for each DAG) like:
record_count, namespace, concept ID, ancestor ID
where the GO IDs are foreign keys. The record_count
might be useful to identify the order of discovery during
traversal.
It might also be useful to include a "generation offset" from
the concept to each ancestor.
Query resolution would then require simple SQL requests
for concept-ancestor pairs.
43
Merging and building ontologies
If you confront 2 databases each with its own ontology,
you MIGHT be able to map one to the other if you want
to combine them or query both using the ontology of
just 1.
There has been a lot off research in this area, and
apparently a handful (or 2) of tools have been
developed to help, but . . .
“There are multiple tools to merge or map ontologies,
but they are quite difficult to use and require some user
editing in order to obtain reliable results.” (Pasquier,
2008)
In fact, there are also “no standardized methods for
building ontologies” (Sevens, et al., 2003), and even
though there exist multiple toolsets to help, building
ontologies remains difficult.
44
Shirky (2008) argues that ontologies are not necessarily
the best way to annotate all kinds of data.
He provides a list of domain and user characteristics
that bode well for success:
“Domain has a small corpus, formal categories,
stable entities, restricted entities, and clear edges.”
The “participants” are expert catalogers and
include authoritative sources of judgment, and
the users are organized, and expert in their
use of the ontology.
He sites the psychiatric Diagnostic and Statistical
Manual (DSM-IV) and the Periodic Table as examples.
One might add that the domain categories are stable,
but not too stable; the categorization structure is
irregular; and there are storage space constraints.
45
Summary
There are many (not entirely consistent) definitions of
ontology.
The “Big” definition provides a concrete toehold that
helps clarify the other definitions, and can be used to
structure further work.
The Gene Ontology is, and can be, used to annotate
life-sciences data.
Programs can be written to use the Gene Ontology and
annotated data to answer queries (prove assertions).
Building ontologies may be difficult, but should be worth
the effort in many circumstances.
46
References
Aktas, Mehmet, and Malon Pierce, Semantic Web and RDF Ontologies.
http://grids.ucs.indiana.edu/ptliupages/presentations/SemanticWeb&RDFOntology.ppt
Berners-Lee, Tim, James Hendler and Ora Lassila, The Semantic Web, Scientific American, May
2001.
http://www.sciam.com/article.cfm?id=the-semantic-web
Blake, Judith, “Using the Gene Ontology for Data Analysis”.
http://www.geneontology.org/teaching_resources/presentations/2004-11_dataanalysis_jblake.ppt
Feigenbaum,Lee, Ivan Herman, Tonya Hongsermeier, Eric Neumann and Susie Stephens, The
Semantic Web in Action, Scientific American, 2007. Ignore this article.
http://thefigtrees.net/lee/sw/sciam/semantic-web-in-action#single-page
Gruber, Tom, “What is an Ontology?”, Personal web site.
http://www.ksl.stanford.edu/kst/what-is-an-ontology.html
Jonquet, Clement, Mark A. Musen, Nigam H. Shah, Help will be provided for this task: OntologyBased Annotator Web Service, International Semantic Web Conference (ISWC08), Karlsruhe,
Germany. May 2008.
http://bmir.stanford.edu/file_asset/index.php/1321/ISWC08_Jonquet_Musen_Shah_final.pdf
Niepert, Mathias, Cameron Buckner, and Colin Allen, Answer Set Programming on Expert
Feedback to Populate and Extend Dynamic Ontologies, Association for the Advancement of
Artificial Intelligence, 2008.
http://inpho.cogs.indiana.edu/Papers/2008-InPhO-flairs.pdf
47
Pidcock, Woody, What are the differences between a vocabulary, a taxonomy, a thesaurus, an
ontology, and a meta-model?, Web article, 2003.
http://www.metamodel.com/article.php?story=20030115211223271
Rubin, Daniel L.,1 Dilvan A. Moreira, Pradip P. Kanjamala, and Mark A. Musen, BioPortal: A Web
Portal to Biomedical Ontologies, AAAI Spring Symposium Series, Symbiotic Relationships
between Semantic Web and Knowledge Engineering, Stanford University, (in press). Published
2008.
http://bmir.stanford.edu/file_asset/index.php/1298/AAAI-BioPortal-2008.pdf
Shadbolt, Nigel, Wendy Hall and Tim Berners-Lee, The Semantic Web Revisited, IEEE
INTELLIGENT SYSTEMS, 2006.
http://eprints.ecs.soton.ac.uk/12614/1/Semantic_Web_Revisted.pdf
Sheth, Amit, Cartic Ramakrishnan, Semantic (Web) Technology In Action:
Ontology Driven Information Systems for Search, Integration and Analysis, IEEE Data
Engineering Bulletin, Special issue on Making the Semantic Web, Real, U. Dayal, H. Kuno, and
K. Wilkinson, Eds., December 2003.
http://lsdis.cs.uga.edu/library/download/SR03-BW.pdf
Shirky, Clay, “Ontology is overrated”, Personal website
http://www.shirky.com/writings/ontology_overrated.html
Stevens, Robert, Carole A. Goble, and Sean Bechhofer, Ontology-based knowledge
representation for bioinformatics, Briefings in Bioinformatics, November 2000.
http://bib.oxfordjournals.org/cgi/reprint/1/4/398?ck=nck
Wikipedia, Ontology (information science), June 2008.
http://en.wikipedia.org/wiki/Ontology_(computer_science)
48