PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
J Biomed Inform. Author manuscript; available in PMC 2012 February 1.
Published in final edited form as:
PMCID: PMC2910209
NIHMSID: NIHMS179365

Cross-Product Extensions of the Gene Ontology

Abstract

The Gene Ontology (GO) consists of nearly 30,000 classes for describing the activities and locations of gene products. Manual maintenance of an ontology of this size is a considerable effort, and errors and inconsistencies inevitably arise. Reasoners can be used to assist with ontology development, automatically placing classes in a subsumption hierarchy based on their properties. However, the historic lack of computable definitions within the GO has prevented the user of these tools.

In this paper we present preliminary results of an ongoing effort to normalize the GO by explicitly stating the definitions of compositional classes in a form that can be used by reasoners. These definitions are partitioned into mutually exclusive cross-product sets, many of which reference other OBO Foundry candidate ontologies for chemical entities, proteins, biological qualities and anatomical entities. Using these logical definitions we are gradually beginning to automate many aspects of ontology development, detecting errors and filling in missing relationships. These definitions also enhance the GO by weaving it into the fabric of a wider collection of interoperating ontologies, increasing opportunities for data integration and enhancing genomic analyses.

Introduction

The Gene Ontology (GO)1 was conceived of as a means of providing structured annotations for genes and gene products, in terms of molecular function (MF), biological process (BP) and cellular component (CC). The current version of the GO has nearly 30,000 classes and 51,000 relationships. As the GO evolves, the relational graph becomes more tangled, which poses a problem for ontology maintenance and visualization. It has long been recognized that a normalized approach to ontology development helps with re-use, maintainability and evolution2,3,4. The OBO Foundry5 was initiated in part to provide a means of normalizing the GO, such that for example the GO definition of “oocyte differentiation” could reference the class “oocyte” in the OBO Cell ontology (CL), and an automated reasoner tool could be used to classify this as a kind of “germ cell differentiation”, based on the CL classification. This is also an example of a ‘re-use’ pattern, common in software engineering.

Almost all of the classes in the GO have textual definitions, crafted for the human users of the GO, but opaque and meaningless to computers, without the use of sophisticated natural language processing. Here we present the results of an ongoing project to render these textual definitions in a computable form as a collection of cross-products, weaving together multiple ontologies. This allows the use of reasoners to automate the more tedious and error-prone aspects of ontology maintenance. We can also use these computable definitions to make cross-ontology queries and better visualize the ontology.

Results

Formalism: Logical Definitions and Cross Products

We provide computable logical definitions for classes using genus-differentia constructs, of the form “an X is a G that D”. Here X is the class we are defining, G is the genus (more general class), and D is the differentia, a collection of characteristics that serve to discriminate instances of X from other instances of G. The differentiae are specified as relationships to other classes, using relations from the Relation Ontology (RO)6 as well as proposed extensions to the RO. In OBO Format (the native means of representing the GO) these are specified using intersection_of tags, which list the necessary and sufficient conditions for a class. For example:

[Term]

id: GO:0032543

name: mitochondrial translation

intersection_of: GO:0006412 ! translation

intersection_of: occurs_in GO:0005739 ! mitochondrion

In OWL Manchester Syntax, this is written as an equivalence axiom between the class mitochondrial translation and the description translation and occurs_in some mitochondrion (see Table 1). In this particular case, the occurs_in relation is not defined in the RO, but we can define this using two existing relations: has_participant and located_in1, such that if a process occurs_in a component, then all the participants of that process are located in that component for the duration of that process.

Table 1
Translation of logical definition section of OBO stanza into OWL and predicate logic. The ir function maps an OBO type level relation to a corresponding instance level relation

In the example above, the logical definition for the process references a GO cellular component class. Often we will want to reference other OBO ontologies, and this introduces multiple dependencies. We therefore partition the full set of logical definitions for GO into cross-product mapping. A cross-product of two ontologies A x B is the set of biologically meaningful classes that can be constructed by extending A using classes from B as differentia. The GO class in the example above would be mapped to a definition in the BP x CC cross product. We distinguish between internal (intra-GO) cross products and external (referencing classes outside GO) cross products, the latter consisting of logical definitions that reference an OBO ontology that is not one of the three ontologies that constitute the GO.

A subset of the intra-GO cross products are the self-cross products: terms that can be defined solely by using terms in the same ontology, such as for example BP x BP.

We use the following set of OBO Foundry candidate ontologies for the cross-products:

  • BP – GO Biological Process
  • CC – GO Cellular Component
  • MF – GO Molecular Function
  • CL – Cell
  • CHEBI – Chemical Entities of Biological Interest
  • PRO – Protein
  • SO – Sequence Ontology[REF]
  • Uberon – Multi-species metazoan anatomy ontology
  • PO – Plant anatomy ontology
  • PATO - qu0alities

Each cross-product mapping is maintained as an individual resource, independent of the others, and each is currently available as optional add-ons to the GO.

We took this formalism and the above ontologies, and using a combination of automated tools and human curation, we have specified logical definition for a large number of classes in the Gene Ontology. Table 2 summarizes the collection of cross-product sets, and table 3 specifies the set of relations used to create these logical definitions.

Table 2
GO logical definitions are partitioned into mutually exclusive cross-product sets. Examples are shown from each of the sets. The second column shows the number of existing GO classes that have been mapped to a logical definition in each set
Table 3
Relations used in logical definitions for biological process classes that reference a physical entity class. For most relations, the instance level relation (bold font) is specified. We assume a standard all-some semantics for the corresponding type-level ...

Location of processes in a cell

The BP x CC cross-product set includes definitions for biological process classes that use cellular component classes as differentiae. Many of these classes require a cellular component class to indicate where the process takes place in the cell, and in this case we use an occurs_in relation – for example, to define mitochondrial translation as equivalent to translation and occurs_in a mitochondrion. We also use this relation in the BP x CL set – for example, we define “oocyte axis specification” as equivalent to “axis specification” and occurs_in “oocyte”.

The occurs_in relation can be defined in terms of existing RO relations – a process occurs_in an entity if all the participants in a process are located in that entity of the duration of that process (see table 3).

Transport and localization

The GO has a rich set of subcellular transport and localization classes. A transport process can have multiple participants and associated locations, we have specified five transport-oriented relations for connecting the transport process to physical entities. A GO class such as vesicle export from nucleus would be defined using the differentiae results_in_transport_of vesicle, results_in_transport_from nucleus and results_in_transport_to cytoplasm. Note that whilst the formal definitions of these relations require the full expressive power of predicate logic, many applications can effectively treat the relations as opaque.

Note that not all transport classes fall strictly into the BP x CC set. Sometimes cell types are referenced, sometimes chemical entity types – in these cases they would fall into the BP x CL or BP x CHEBI set.

Inputs and outputs of processes

Other classes represent the formation or dissociation of cellular components, and here we use has_input and has_output relations – for example, “spindle assembly” is defined as “cellular component assembly” and has_input “spindle”. Both these are sub-relations of the has_participant relation defined in RO. Note that it is not sufficient to use the has_participant relation alone, because we need to distinguish the role or function played by the participant in the process.

BP x BP subset

Many BP classes can be defined using a BP class in the differentia. These definitions are grouped into the BP x BP set, and the vast majority of these use the part_of relation. For example, the different phases of the cell cycle can be subtyped according to whether they are part of the mitotic or meiotic cell cycle i.e. the cross-product of the sets { G1 phase, G2 phase, S phase, interphase, …} x { meiotic cell cycle, mitotic cell cycle } using the part_of relation, yielding logical definitions for classes such as “G1 phase of mitotic cell cycle”.

Biological regulation

GO includes 3 broad categories of regulatory processes – regulation of molecular function, regulation of biological process, and regulation of biological quality – these comprise 3 distinct cross-product sets. The first two are intra-GO; the latter comprise classes such as ‘regulation of cell volume’ where ‘cell volume’ can be defined using a combination of the PATO ontology of biological qualities7, together with anatomical ontologies.

The cross products make use of 3 new relations introduced into GO – regulates, negatively_regulates and positively_regulates. We define the regulates relation in terms of qualities – a regulatory process results in the change in magnitude to some quality (for example, the width of or pressure in an artery). This change in quality has an effect on the frequency, rate or duration of some other type of process. If this results in an increase, then we use positively regulates, and for the converse, negatively regulates.

We have created a separate cross-product set for the more complex multi-organism interaction regulation classes such as “modulation of intracellular transport in other organism during symbiotic interaction”. The logical definitions we provide here are necessarily a simplification, as we must go beyond the current expressive capabilities of OBO or OWL in order to represent interactions between organisms.

Gross Anatomy

GO has many classes that reference the various parts of multicellular organisms. This includes both developmental classes such as “lung development” and “heart looping” as well as physiological classes such as “muscle contraction”.

Gross anatomy proves a challenge because GO is applicable across all forms of life and the main OBO gross anatomy ontologies are specific to a species or taxon. Here we decided to use Uberon, a multi-species metazoan “uber”-anatomy ontology, recently constructed for the purposes of comparing phenotypes across species8. We extended Uberon by extracting the implicit anatomical ontology embedded in the GO. Uberon is used in the definition of classes such as “neural tube formation”, which is defined as the intersection of “anatomical structure formation” and results_in_formation_of “neural tube”. These definitions are part of the BP x Uberon set. Uberon covers only animals; plant development classes are in BP x PO (plant anatomy ontology). There are also individual species-specific extensions such as BP x Fly_anatomy and BP x ZFA (Zebrafish anatomy), in which terms such as “haltere disc development” and “pectoral fin development” are defined. If a GO class is not applicable beyond a specific taxon or represented taxonomic group, then the species-centric anatomy ontology is used.

Assigning necessary and sufficient conditions to specific developmental classes is not always trivial, as in the case of “heart looping”. Here we are currently content to logically define the parent class “heart morphogenesis” and leave the more specific class with a logically imprecise textual definition.

Cell types

GO covers the development and formation of structures at different levels of organization, from subcellular components through to whole organisms. We use the species-neutral OBO Cell ontology (CL) 9 for defining classes such as “oocyte differentiation” in the BP x CL set.

Molecules and proteins

Molecular and chemical entities are represented in the CHEBI ontology10, with proteins represented in PRO11. We use these in 3 cross-product sets, {BP,MF} x CHEBI and BP x PRO. The CHEBI sets have been described previously12.

CHEBI is an important ontology from the point of view of GO, as the majority of molecular functions and a substantial portion of cellular processes can be defined by referencing chemical entity classes using the has_input and has_output relations. Molecular function classes such as ‘lactase activity’ can be defined as catalyzing the reaction with inputs lactose, water, and outputs D-glucose and D-galactose. Metabolic processes can also be defined using input and output relations and a genus class that specifies whether catabolism or biosynthesis is occurring – for example ‘maltose biosynthetic process’ as equivalent to the intersection of ‘biosynthetic process’ and has_output ‘maltose’. Chemical entities can also play the role of cargo in transport processes.

The Protein Ontology is still relatively new, so this last set is currently relatively small. We also intend to work with the PRO curators to make a CC x PRO set, in which generic protein complexes such as “core TFIIH complex” are defined in terms of their constituent generic protein parts.

We used the Sequence Ontology (SO) 13 for defining a small subset of classes such as “group I intron processing”. Initially the exact meaning of these definitions were unclear, because the term “intron” in SO was previously used in the sense of both the molecule and the genomic sequence that could give rise to that molecule. This issue is currently being addressed in SO, and we will likely use the new Sequence Molecule Ontology (SOM) for defining the GO classes14. We also anticipate the need to use the new RNA Ontology15 for a small subset of GO classes.

Spatial relations for cellular components

Many of the classes in CC can be assigned logical definitions based on parthood relations to other components – for example, “nuclear chromosome” is a chromosome that is part_of a nucleus. We also use the reciprocal relation has_part to define classes such as microtubule skeleton as equivalent to a cytoskeleton and has_part microtubule. For other definitions in CC x CC, we introduce additional spatial relations, such as surrounds and surrounded by. The surrounds relation is used to define membranes and envelopes, and the surrounded_by relation is used to define lumen classes. There are additional spatial relations such as continuous_with, spans and adjacent_to. The names and logical definitions of these relations are still in progress, so we do not include a detailed summary of them here, but this will be available in a future publication.

The GO CC ontology has many classes representing complexes, some of which are defined by their constituent parts, others by function. The former will have logical definitions in the previously mentioned CC x PRO subset, and the latter in the CC x MF cross-product set.

Some cell component classes are differentiated by the cell type of which they are a part – for example, a sarcoplasm is a cytoplasm that is part_of a muscle cell. We place GO classes such as sarcoplasm to the CC x CL set, most of which use the part_of relation. For others, such as neuromuscular junction we use adjacency relations (i.e. adjacent_to motor neuron axon and contractile fiber).

Use of reasoners as an ontology maintenance tool

One of the goals of assigning computable logical definitions is to leverage tools such as reasoners to automate tedious and error-prone aspects of ontology maintenance. The current set of logical definitions can be used by a variety of different reasoners - we use the OBO-Edit16 reasoner because it is integrated within the normal editing environment for the GO and provides incremental reasoning support.

We have not found any reasoner that is capable of reasoning over the union of the GO plus all cross-product sets plus all referenced ontologies. However, we are able to reason over individual cross-product sets and their referenced ontologies individually.

We use reasoning primarily for ontology maintenance, to compute and check the subsumption hierarchy. We use the reasoner in ‘repair mode’ – we invoke the reasoner to spot mistakes and fill in missing links in the ontology, always asserting links that can be automatically computed, provided the supporting links are valid (editors frequently spot biologically invalid inferences, but these are necessarily due to upstream errors, which are corrected before asserting implied links). This ensures that editors can edit the ontology without invoking the reasoner over the union of all logical definitions, which may be time-consuming. This stands in contrast to how the reasoner is used in SO and the fly anatomy ontology (FBbt), in which the asserted hierarchy is minimal and computable links are calculated dynamically (except in the export version of these ontologies). For example, in the editorial version of the fly anatomy there is no asserted link between “olfactory neuron” and “sensory neuron” – this is computed by the reasoner as part of the release process, and visible to the majority of ontology consumers who require the full subsumption graph. The full methodology is described in a tutorial (http://www.bioontology.org/wiki/index.php/OBO-Edit).

The GO ‘biological regulation’ hierarchy in particular has benefited from this work, with over 2000 missing links added to GO, and the reasoner used for all new additions (Figure 1)

Figure 1
Using logical definitions to infer placement in hierarchy. Boxes denote classes, and edges denote relationships between classes. is_a relations are depicted as [I], regulates links as [R]. Bold text denotes the class has a logical definition in the regulation ...

Discussion

The challenges of coordinated ontology development

In a traditional ontology development scenario, reasoners may be used to infer the subsumption hierarchy of composite classes based on properties described in terms of simpler classes – for example, inferring that carbohydrate metabolism subsumes glucose metabolism based on the fact that the carbohydrate class subsumes the glucose class. Because we are working in an environment that involves multiple ontologies we have found it useful to also perform a form of inverse-reasoning or abductive reasoning, in which we make inferences from the GO to referenced ontologies such as the CL or CHEBI17. This is useful for finding inconsistencies between GO and other ontologies, and within the GO. Using this method we have uncovered a number of fundamental differences between classification in CHEBI and the implicit chemical entity ontology in GO. For example, the CHEBI definition of “carbohydrate” is very non-specific and inclusive, and includes “nucleotide” and even “DNA” as subtypes. When we use the BP x CHEBI logical definitions to fill in missing relationships in GO we end up classifying “nucleotide biosynthesis” as a subtype of “carbohydrate biosynthesis”. This turned out to be very unintuitive for all the GO ontology developers, and on further investigation and discussion with the CHEBI team it transpired that even for such a basic term as “carbohydrate” different communities had different definitions. This issue has not yet been satisfactorily resolved and proves to be a stumbling block on the adoption to a more automated approach to ontology maintenance.

Future applications

The primary benefits of assigning logical definitions are in automating ontology maintenance. We are exploring a number of additional applications of these cross-products beyond this use case.

One application is aligning the GO with pathway databases and systems biology resources. Databases such as Reactome model events in terms of their inputs and outputs, which are annotated with CHEBI classes. The locations of inputs as are annotated with GO CC classes. We can use our logical definitions to query pathway databases using the necessary and sufficient conditions for a class, and then infer subsumption between GO classes and pathway events. We can then automatically propagate the gene products which actively participate in this event to the GO class, generating new annotations in a way that is reliable and synchronized with pathway databases.

We can also use the internal cross-product logical definitions for logical and probabilistic inference of annotations. For example, given that a gene product is annotated to ‘translation’, and that gene product has been shown in different experiments to be localized to ‘mitochondrion’, but never outside the mitochondrion, then there is a reasonable chance that this participates specifically in ‘mitochondrial translation’. This can be expressed more generally as: if G actively participates in P and is localized to C, and never to anything that is not part_of C, then G probably participates in P’, where P’ = P and occurs_in C. In order to express these kinds of rules we must move beyond strict deductive formalisms. We can make the converse deduction with more confidence – if a gene product actively participates in ‘mitochondrial translation’, we can prove by the definition of the occurs_in relation that the gene product is localized to the mitochondrion.

The cross-products can also potentially be used for visualization. As the GO expands, the average number of paths-to-root per class increases, which proves problematic for standard graph-based visualization techniques, where the graphs begin to resemble tangled hairballs. The cross-product definition makes the construction of the class explicit and allows for disentangled displays. For example, rather than showing a tangled graph for ‘cysteine biosynthesis’, it is possible to show two graphs, one for ‘biosynthesis’ and another for ‘cysteine’, communicating the same information but in a more compact understandable fashion.

One of the most common uses of the GO is for class-enrichment analyses, in which a gene set (for example, a set of genes co-expressed under some condition) is statistically analyzed to find over-represented classes. A rate limiting step here is the fixed structure of pre-composed classes in the GO. For example, serotonin transmission and serotonin biosynthesis are in different unrelated parts of the GO graph. If a set of genes is involved in a mix of these two classes, then the p-value for each will be weaker for a p-value for ‘serotonin’, if BP x CHEBI and CHEBI are used in the analysis. The GO-CHEBI cross-products set has already been applied in a similar way, to augment inference of biological pathways from gene expression data18.

We have also used the logical definitions the integration of phenotype ontologies. Sometimes phenotype descriptions are anatomical, and refer to entities represented in anatomical ontologies such as lungs, hearts and so on. Other phenotype descriptions are developmental, and refer to processes represented in BP, such as lung development and heart formation. Using the BP x Anatomy subset we can reconcile these descriptions such that they can be used in reasoning19.

On Pre and post-composition

The GO does not pre-compose classes for all biologically meaningful compositions of classes, as this would lead to a large, unwieldy ontology. The guiding principle is to generate compositional classes where the differentiae are important to the biology. For example, there is a strong argument for pre-composing ‘mitochondrial translation’ as the mechanism of translation in the mitochondria is different than that elsewhere. This choice can be justified by the appearance of ‘mitochondrial translation’ in gene set enrichment analysis results. In contrast, it is unlikely GO would pre-compose ‘ossification of middle phalanx of left little finger’, as there is nothing to distinguish this kind of ossification from other kinds of middle phalanx ossification other than trivially by location. It also seems unlikely we would see a class like this show up in class enrichment analyses (unless of course this was an unlikely gene expression study focusing on this body part).

It would be possible to devise an annotation system whereby annotators could specify class expressions such as ‘translation and part_of mitochondrion’ and avoid pre-composing this class in the GO, allowing for a more minimal ontology which would be easier to maintain to explore. However, this would be a mixed blessing for annotators, as they would have to understand the additional expressivity, and post-composition may be more time consuming. In addition, the added complexity would propagate to databases and tools that consume the GO in any way. For example, most databases and tools now store annotations as pairwise associations between a pair of IDs. For a database to support post-composition, it would have to add additional description-logic style structures which would be difficult to query over in traditional systems.

We are exploring an approach whereby annotators can extend GO classes on-the-fly, i.e. selecting compositions from the cross-product at annotation time. For example, an annotator can select the GO class ‘mitochondrial membrane’ for a cellular component annotation and extend this using a differentia ‘part_of Purkinje cell’, with the differentia class coming from CL. This is logically equivalent to annotating to a class ‘mitochondrial membrane of Purkinje cell’, but avoids bloating the ontology with the full set of biologically valid classes in the CC x CL cross-product.

Future development

The results in this paper describe a work in-progress. We have many remaining classes in the GO for which the specification of logical definitions would be relatively simple and beneficial. In many cases the hold-up has been the bottleneck in generating the required classes in external ontologies. However, the situation is improving here. The rate of addition of new classes to the CL was previously a problem, but this ontology is now being actively developed20. The results in this paper were generated when PRO was a relatively new ontology, lacking many of the classes required to flesh out parts of GO under “protein binding”, “protein biosynthesis” and so on. However, in January 2010 there was a new release of the PRO with a far more comprehensive coverage of the proteome referenced in GO (although at this time it is still lacking many of the abstracted classes – for example it has various different types of E3 protein ligase, but not the generic class “E3 protein ligase”). We expect our coverage of protein-related GO classes to expand dramatically.

However, the GO is constantly being developed and expanded. It has become clear that the retrospective approach of assigning logical definitions post-hoc is not sustainable, and we are therefore moving to a more prospective approach. As of January 2010, the GO introduced a policy where all new regulation classes have logical definitions assigned first, which greatly increases maintainability.

Conclusions

The extended collection of cross-product resources described here represents a significant advance in the evolution of the GO and its integration with other OBO ontologies. The use of these logical definitions, in conjunction with a reasoner has substantially increased the quality of the GO and eased the more prosaic aspects of ontology maintenance. We are exploring applications beyond ontology automation, in particular to increase annotation coverage in GO through alignment with pathway databases, and through probabilistic inference using the logical definitions.

This work also highlights the importance and necessity of the OBO Foundry effort, particularly with respect to efforts to create single orthogonal well-partitioned ontologies each representing a distinct domain of biology.

Acknowledgments

This work is supported by the NHGRI, via the Gene Ontology Consortium, HG002273.

Footnotes

1We follow the conventions of Smith et al 2005 and use italics to denote relations between types and bold font to denote relations between instances.

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Methods and availability

In contrast to some ontology development efforts, in which computable definitions are assigned when classes are created, we have been working retrospectively, constructing logical descriptions for pre-existing classes. To help us with this task we use Obol21, which heuristically generates proposed logical definitions based using ontology-specific grammars. Ontology editors then vet the definitions, often substantially.

The full extended GO can be obtained on the GO wiki: http://wiki.geneontology.org/index.php/Category:Cross_Products

Comments and contributions are welcome.

References

1. Ashburner M, Ball CA, Blake J, Butler H, Cherry J, Corradi J, Dolinski K, Eppig J, Harris M, Hill D, Lewis S, Marshall B, Mungall C, Reiser L, Rhee S, Richardson J, Richter J, Ringwald M, Rubin G, Sherlock G, Yoon J. Creating the gene ontology resource: design and implementation. Genome Res. 2001;11:1425–1433. [PubMed]
2. Hill DP, Blake JA, Richardson JE, Ringwald M. Extension and integration of the gene ontology (GO): combining GO vocabularies with external vocabularies. Genome Res. 2002;12:1982–91. [PubMed]
3. Wroe CJ, Stevens R, Goble CA, Ashburner M. A methodology to migrate the gene ontology to a description logic environment using DAML+OIL. Pac Symp Biocomput. 2003:624–35. [PubMed]
4. Rector AL. Modularisation of domain ontologies implemented in description logics and related formalisms including OWL; Proceedings of the 2nd international conference on Knowledge capture, ACM; 2003.pp. 121–128.
5. Smith B, Ashburner M, Rosse C, Bard J, Bug W, Ceusters W, Goldberg LJ, Eilbeck K, Ireland A, Mungall CJ, Consortium TO, Leontis N, Rocca-Serra P, Ruttenberg A, Sansone S-A, Scheuermann RH, Shah N, Whetzel PL, Lewis S. The OBO Foundry: coordinated evolution of ontologies to support biomedical data integration. Nat Biotechnol. 2007;25:1251–1255. [PMC free article] [PubMed]
6. Smith B, Ceusters W, Klagges B, Kohler J, Kumar A, Lomax J, Mungall C, Neuhaus F, Rector AL, Rosse C. Relations in biomedical ontologies. Genome biology. 2005;6(5):R46. [PMC free article] [PubMed]
7. Gkoutos GV, Green EC, Mallon AM, Hancock JM, Davidson D. Using ontologies to describe mouse phenotypes. Genome Biol. 2005;6:R8. [PMC free article] [PubMed]
8. Washington Nicole L, Haendel Melissa A, Mungall Christopher J, Ashburner Michael, Westerfield Monte, Lewis Suzanna E. Linking Human Diseases to Animal Models using Ontology-based Phenotype Annotation. PLoS Biology. 2009;7(11) [PMC free article] [PubMed]
9. Bard J, Rhee SY, Ashburner M. An ontology for cell types. Genome Biol. 2005;6:R21. [PMC free article] [PubMed]
10. Degtyarenko K, de Matos P, Ennis M, Hastings J, Zbinden M, McNaught A, Alcantara R, Darsow M, Guedj M, Ashburner M. ChEBI: a database and ontology for chemical entities of biological interest. Nucleic acids research. 2008;36:D344–350. Database issue. [PMC free article] [PubMed]
11. Natale DA, Arighi CN, Barker WC, Blake J, Chang TC, Hu Z, Liu H, Smith B, Wu CH. Framework for a protein ontology. BMC bioinformatics. 2007;8(Suppl 9):S1. [PMC free article] [PubMed]
12. Bada M, Hunter L. Identification of OBO nonalignments and its implications for OBO enrichment. Bioinformatics. 2008;24:1448–1455. [PubMed]
13. Eilbeck K, Lewis SE, Mungall CJ, Yandell MD, Stein LD, Durbin R, Ashburner M. The Sequence Ontology: a tool for the unification of genome annotations. Genome Biology. 2005;6(5) [PMC free article] [PubMed]
14. Mungall CJ, Batchelor C, Eilbeck K. Evolution of the Sequence Ontology terms and relationships. JBI (this issue) [PMC free article] [PubMed]
15. Batchelor Colin, Bittner Thomas, Eilbeck Karen, Mungall Chris, Richardson Jane, Knight Rob, Stombaugh Jesse, Zirbel Craig, Westhof Eric, Leontis Neocles. The RNA ontology (RNAO): An ontology for integrating RNA sequence and structure data. JBI (this issue)
16. Day-Richter J, Harris MA, Haendel M, Lewis S. OBO-Edit--an ontology editor for biologists. Bioinformatics. 2007;23:2198–2200. [PubMed]
17. Bada M, Mungall C, Hunter L. A Call for an Abductive Reasoning Feature in OWL-Reasoning Tools toward Ontology Quality Control; 5th OWL Experiences and Directions Workshop (OWLED 2008); 2008.
18. Tipney HJ, Leach SM, Feng W, Spritz R, Williams T, Hunter L. Leveraging existing biological knowledge in the identification of candidate genes for facial dysmorphology. BMC Bioinformatics. 2009;10(Suppl 2):S12. [PMC free article] [PubMed]
19. Mungall Christopher, Gkoutos Georgios, Smith Cynthia, Haendel Melissa, Lewis Suzanna, Ashburner Michael. Integrating phenotype ontologies across multiple species. Genome Biology. 2010;11(1):R2. [PMC free article] [PubMed]
20. Diehl Alexander D., Augustine Alison Deckhut, Blake Judith A., Cowell Lindsay G., Gold Elizabeth S., Gondré-Lewis Timothy A., Masci Anna Maria, Meehan Terrence F., Morel Penelope A., NIAID Cell Ontology Working Group. Nijnik Anastasia, Peters Bjoern, Pulendran Bali, Scheuerman Richard H., Yao Q. Alison, Zand Martin S., Mungall Christopher J. Hematopoietic Cell Types: Prototype for a Revised Cell Ontology. Journal of Biomedical Informatics (this issue) [PMC free article] [PubMed]
21. Mungall CJ. Obol: Integrating Language and Meaning in Bio-Ontologies. Comparative and Functional Genomics. 2004;5(7):509–520. [PMC free article] [PubMed]