Increases in the size and scope of the clinical measurement ontology

As shown in Table 2 and Figure 1A, the scope has expanded from 13 direct subclasses of the term “clinical measurement” to 26. The new branches include coverage for additional organ systems (alimentary/gastrointestinal system, endocrine/exocrine system, immune system, musculoskeletal system, nervous system, and skin) as well as coverage for measurements that do not necessarily come to mind as “phenotypes”. These include branches related to disease population measurements such as incidence and prevalence, to disease processes such as onset/diagnosis and progression, and to mortality and survival. In these cases, quantitative values are commonly assessed and reported in the literature. For example, researchers will give a number for the percentage of a study population that develop a disease within a given period of time, report the age at which a disease state is detected, or track the proportion of animals in a study population which are surviving at a series of time points.

	Original branches of the CMO:	New branches added to the CMO:
Organ systems:
	Cardiovascular system	Alimentary/gastrointestinal system
	Liver/biliary system	Endocrine/exocrine system
	Renal/urinary system	Musculoskeletal system
	Reproduction	Nervous system
	Respiratory system	Skin
		Immune system
Other branches:
	Blood measurement	Body movement measurement
	Body morphological measurement	Chemical response/sensitivity measurement
	Body temperature	Disease population measurement (incidence/prevalence)
	Cell measurement	Disease process measurement (onset/diagnosis, progression, severity)
	Consumption measurement	Exudate measurement
	Growth measurement	Mortality/survival measurement
	Tissue composition measurement	Organ measurement
	Tumor measurement

Table 2 lists all of the direct children of the parent CMO term “clinical measurement” as of July 2013, v2.5. These are divided into the list of terms that existed at the time of the original publication of the ontology on the left and those which have subsequently been added to the ontology on the right. Increases in the scope of the CMO have almost doubled the number of direct subclasses of the parent.

A branch for chemical response and sensitivity measurements which covers the results of a variety of both in vivo and ex vivo drug and chemical tests was added, as was a branch for “exudate measurements” for use with measurements made on extravasated fluid or other substances.

The term originally labeled as “organ measurement” (CMO:0000068) was changed to “organ morphological measurement” in keeping with its placement under “body morphological measurement” and above terms which described only organ morphology. A new term for “organ measurement” (CMO:0000669) was created directly under the root term “clinical measurement” and linked as a parent to “organ morphological measurement” (CMO:0000068) (Figure 1B). These changes have allowed inclusion of physiological measurements related to the specified organs in addition to their corresponding morphological measurements.

A new branch for “body movement measurement” addresses a common area of study in rodent research that was not covered in the earlier version of the CMO. This branch is designed to include both involuntary movements, such as measurement of the acoustic startle response (CMO:0001519), and voluntary movements. In the rodent research literature, measurements of voluntary movements such as locomotor behavior in an open field apparatus, rearing, or freezing are often presented as measurements of the emotional state of the animal (e.g., anxiety [11]). Although this is a common interpretation of the results, the psychological state is not the actual quantity being measured. Additionally, such movement measurements can be used in other contexts. A cursory search of the rat literature resulted in articles in which movement in an open field apparatus was used to assess learning/memory [12], ethanol-related hyperactivity [13], the sedative effects of drug treatments [14], the locomotor effects of vestibular dysfunction [15], and the effects of cholinergic denervation of the hippocampus [16]. This being the case, the branch was developed as representing measurements of movement in general, not of psychology or emotionality. Also, because the same measurements are made across a number of different types of apparatus, the specifics of the apparatus are assigned via the MMO rather than being included in the CMO terms.

Collaboration with the Animal QTL Database (QTLdb) [17, 18] has led to the addition of a substantial number of CMO terms related to agricultural animal assessments. These include terms for measurements commonly used by the agricultural community to assess the composition and yield of milk for cattle and sheep, as well as measurements of fowl eggs, of fat and muscle morphology and fat composition in cattle and pigs, and of feed intake and weight gain in cattle, pigs, sheep, and chickens.

Increases in the size and scope of the measurement method ontology

The Measurement Method Ontology covers the domain of the specific methods used to make the measurements represented by the CMO terms, i.e., “how it was measured”. This being the case, development of this ontology is closely coordinated with the development of the CMO and it has likewise increased in size and scope [19]. The number of terms in the MMO has increased from 195 to 402, the maximum depth of the ontology has increased to 8, the percentage of classes with two or more parents has risen to 7.2%, the percentage of classes with a single subclass is 14.9%, and the average branching factor is 0.93 (Table 1). The MMO is subdivided into two major branches: “in vivo method” for methods performed in or on a living body, and “ex vivo method” for procedures performed outside the living body. Improvements to the in vivo branch include the addition of terms for body movement methods, such as subbranches for types of test enclosures, mazes, and treadmills, and addition of more general branches for “flowmetry” and “body fluid collection method”. New subbranches under “ex vivo method” include radioactivity and volume measurement methods, as well as a branch for “isolated cell method” which corresponds to expansion of the CMO “cell measurement” branch. In several cases, what was originally a single term has been expanded into a larger branch. For instance “gel electrophoresis”, originally a direct child of “ex vivo method”, now appears within the “molecular separation method” branch of the MMO (Figure 2).

Increases in the size and scope of the experimental condition ontology

Because incorporation of data from new areas of research requires the addition of new condition terms, the Experimental Condition Ontology has expanded from 110 to 346 terms, the maximum depth of the ontology has now increased to 8, the percentage of classes with two or more parents is 14.2%, the percentage of classes with a single subclass is 17.1% and the average branching factor is 0.87 (Table 1) [20]. New branches under “experimental condition” include “controlled visible light condition”, “controlled in situ organ condition” and “pathogen”. New terms include “sample resting period”, which was necessitated by experiments in which separate measurements were made on a sample before and after the sample was allowed to sit for a specified period of time. The only difference in the conditions between the two values was the “sample resting period”. In addition, a term for “perfusate” was added within the more general “solution” branch to describe experiments performed on isolated organs. The terms “surgical implantation” and “surgical removal” were moved under the new “surgical manipulation” term, and “fasting” was incorporated into the existing “diet” branch.

Consideration was given to simply using the ChEBI ontology for chemical conditions. However, ChEBI is fundamentally an ontology of chemical structures. We would argue that the concept of the use of a chemical as an experimental condition is qualitatively different than that of a chemical as a structure or molecule. In addition, ChEBI is often used in annotation of molecular level gene-chemical interactions which differs from an annotation of a chemical bolus or solution being administered as an experimental stressor. The decision was therefore made to include terms for chemical conditions in the XCO and express the relationship between such a condition and the structure and role of the referenced chemical via cross references to the appropriate ChEBI ID.

Improvements to textual definitions

Work is currently ongoing to both increase the proportion of terms with textual definitions and standardize the format of those definitions for all three ontologies. As shown in Table 1, at the time of the original publication 62% (328/523) of CMO, 59% (116/195) of MMO, and 69% (76/110) of XCO terms had assigned definitions. This proportion has increased to 84% (1427/1691), 81% (326/402), and 92% (320/346), respectively.

As terms are defined, definitions for words or phrases that will be reused are added to a growing list of standardized definition “fragments”. Definitions are currently written manually rather than being automatically generated, but the structure is based on the standard genus-differentia model so that the definition of the child includes the definition of the parent with the addition of applicable differentiating information. As much as possible, each definition is written in such a way that it “stands alone”, that is, so that the user is not required to go up the tree to find the definition of the more general concept. In this way, the definition of “plasma glucose level” (CMO:0000042) has been expanded from “The level of glucose found in a specific volume of plasma” to “Measurement of the amount of glucose, the monosaccharide sugar, C6H12O6, occurring widely in plant and animal tissues which is one of the three dietary monosaccharides that are absorbed directly into the bloodstream during digestion, is the end product of carbohydrate metabolism, and is the chief source of energy for living organisms, in a specified volume of plasma, the fibrinogen-containing fluid portion of the blood in which the particulate components are suspended” in order to incorporate the fragments which define level, glucose, and plasma, respectively. A list of the standard definition fragments currently in use is provided as Additional file 1.

Additional improvements

Additional improvements have been implemented to bring the development of the three ontologies into line with established best practices [23]. Because the development of these ontologies was carried out collaboratively, over time textual information such as definition source Uniform Resource Identifiers (URI) was entered using a variety of formats. For instance, at one point “Dorland’s Illustrated Medical Dictionary, 31st Edition” [24], one of a number of sources used frequently for all three ontologies, was represented by 14 slightly different URIs, most of which differed by as little as the inclusion of a period or apostrophe, or the designation of the edition. Although such differences are simple for the human mind to interpret, they make the information difficult to interpret by parsers and other computer applications. These have all been standardized to “Dorland:Dorlands_Illustrated_Medical_Dictionary--31st_Ed”. As this example also illustrates, definition source URIs have been reformatted to remove all “non-URI” characters as defined by the World Wide Web Consortium (W3C) [25]. According to the W3C document, the characters permitted for a URI which do not have a reserved purpose include upper- and lowercase letters (A-Z/a-z), digits (0–9), hyphen, period, underscore, and tilde. All definition source URIs for the CMO, MMO, and XCO have been reformatted so that only those characters are used. Also, to further increase the standardization and traceability of definitions, the applicable ISBN number has been added to the list of source URIs when a hard-copy book is used rather than an online resource. A representative list of definition source URIs has been provided as Additional file 2.

Finally, a standardized system for file versioning has been implemented, including minor version increments for ongoing term and definition additions and major version increments for global changes to the contents or structure of the ontologies. For example, the standardization of the definition source URIs was considered a global change to the contents of the ontologies and warranted the increase of the major version number for each ontology from 1.x to 2.0. The current version number for each ontology can be found as the “data-version” notation in the header of the ontology file (See Additional file 3). The data version for each file is also given in the list of ontology files available on the applicable NCBO BioPortal ontology page. The version numbers referenced in this paper are v2.5 for the CMO, v2.3 for the MMO, and v3.0 for the XCO.

Expanded applications

RGD has substantially expanded its use of the CMO, MMO, and XCO for integration of complex datasets in the PhenoMiner project. The number of averaged or summary records in the PhenoMiner database has increased from approximately 13,000 in May, 2012 to now over 69,000. This includes incorporation of additional high-throughput data from the PGA and the PhysGen Knockout project for the rat [29] as well as records derived from manual review of the literature to find and integrate quantitative phenotype data. Such data was previously difficult to locate, because it is often dispersed in tables, text, figures, and figure legends incorporated in the body of the paper or included with the supplementary data, and was even more difficult to compare across studies. Easy access to consolidated results across rat strains and experiments is now available in the RGD PhenoMiner tool [30]. Results for a single strain can be accessed directly in PhenoMiner or from the strain phenotype profile section, labelled “Phenotype Values via PhenoMiner”, on the RGD strain report page. All CMO terms for which data exists in the PhenoMiner database are listed and terms link to the corresponding data in the tool display (Figure 3).

High-throughput phenotyping projects such as the PGA often store the quantitative data from each individual rat that is tested. When such data is available each individual measurement is stored separately in the PhenoMiner database, in addition to being grouped and averaged to form a subset of the aforementioned summary records. Currently the number of individual records is over 563,000.

CMO, MMO, XCO, VT, and Rat Strain (RS) ontology terms are also assigned to QTL records at RGD (Figure 4) with over 80% (1578/1911) of RGD’s rat QTLs annotated to date. This provides clear experimental design information to users, allows the user to examine, query, and group data by experimental parameters, and links specific sites on the rat genome to the quantitative measurement data in PhenoMiner.

In addition to extraction of phenotype data from the literature by curators, researchers who carry out phenotyping projects on rat strains are encouraged to submit their data directly to RGD. A submission form has been posted on the RGD website to facilitate the process [31]. Researchers collaborate with RGD staff members during the submission process to ensure that their data is integrated into the resource correctly and in a timely fashion.

Researchers are also encouraged to submit term requests for inclusion in the ontologies and/or to suggest modifications and improvements to the vocabularies. Those wishing to submit such requests and suggestions can do so using the “Contact Us” link at the top of the RGD webpages [1] or the contact information supplied on the appropriate BioPortal ontology description pages [32]. Plans are underway to implement tracking software, such as a SourceForge web page [33], to facilitate this process.

Beyond RGD and the Animal QTLdb, the CMO, MMO, and XCO ontologies and the associated QTL annotation data are being used by researchers and other databases via the freely-accessible RGD ftp site. In the past six months, each of the ontologies has been downloaded from the site between 190 and 314 times (CMO: 314 requests; MMO: 211 requests; XCO: 190 requests). The total number of downloads of the data annotation files were 71 for the CMO, 68 for the MMO, and 67 for the XCO. Analysis of the ftp logs shows that the file requests originate from a variety of institutions including universities, medical schools, government institutes and pharmaceutical companies, and from locations in the United States, Europe and Asia, demonstrating the utility of both the ontologies themselves and the associated annotations.

Although these ontologies were originally designed to be used together, they also have utility individually. One such example was recently demonstrated at the 4th International Conference on Biomedical Ontology where Goldfain et al. presented their work on the use of ontologies to contextualize the measurement of vital signs in individuals [34]. They use a subset of the XCO to incorporate conditions such as “standing position” (XCO:0000083).

For researchers interested in using the ontologies or the associated data, or in submitting their own data for incorporation into the data set, help is available on the RGD website. Recently updated help pages [35] give information on ontologies in general and their use at RGD, as well as detailed instructions on the use of the PhenoMiner tool, the Phenotypes and Models portal, and the QTL report pages. Tutorial videos such as the “Introduction to the RGD Phenotypes and Models Portal” video [36] demonstrate step-by-step the use of specific tools. The RGD “Introduction to Biomedical Ontologies” tutorial series [37] is geared toward the “ontology novice” and gives basic information about what an ontology is and how it might be used. Finally, help is always available by contacting the RGD curators and developers via the “Contact Us” link at the top and bottom of any RGD web page [1].

Future directions

There are continuing efforts to standardize both the structure and content of the ontologies. Future development efforts will include the use of standard ontology tools and semantic reasoners, as well as continuing consultations with domain experts to more systematically fill in gaps. Also, in the early stages of development, a number of instrumentation terms were included as measurement methods, for example, “oral thermometer, digital” (MMO:0000196) [19]. However, this term for an object was an is_a subclass of “thermometry”, a method, creating an obvious problem since an object is not a method. Work is underway to review the ontologies both manually and through the use of reasoners to find and correct these types of logical inconsistencies.

Efforts are also underway to add systematic cross referencing from CMO, MMO and XCO terms to related concepts in other ontologies. As previously mentioned, cross references from XCO terms to ChEBI have already been added. Going forward, we anticipate adding similar cross references to ontologies related to disease, cell types, and anatomy. Results from NCBO’s Mappings tool [38, 39] will be used as a starting point for finding and documenting such inter-ontology relationships. In addition, we are investigating algorithms which may enable us to map between related terms even when the terms do not use identical text.

In keeping with the common practice for ontology development, the terms, class definitions and structure of these ontologies will be reviewed yearly or as needed to ensure that they remain up to date with advances in the associated research domains and that they conform to both initial and newly identified development requirements [40]. As the number of collaborating domain experts for the development of these ontologies grows, regular discussions with those collaborators will be scheduled to review terms, ontology structure and definitions. Our location at the Medical College of Wisconsin is ideal in this respect since MCW houses a large and varied community of basic rat researchers, clinical researchers and clinicians. Such collaborations have already helped us improve the ontologies.

Finally, in order to make these ontologies usable for tools and software designed for OWL-formatted vocabularies, we will make the CMO, MMO and XCO available in the OWL format in the near future.