- The International Centers of Excellence in Malaria Research (ICEMR) projects involve studies to understand the epidemiology and transmission patterns of malaria in different geographic regions. Two major challenges of integrating data across these projects are: (1) standardization of highly heterogeneous epidemiologic data collected by various ICEMR projects; (2) provision of user-friendly search strategies to identify and retrieve information of interest from the very complex ICEMR data. We pursued an ontology-based strategy to address these challenges. We utilized and contributed to the Open Biological and Biomedical Ontologies to generate a consistent semantic representation of three different ICEMR data dictionaries that included ontology term mappings to data fields and allowed values. This semantic representation of ICEMR data served to guide data loading into a relational database and presentation of the data on web pages in the form of search filters that reveal relationships specified in the ontology and the structure of the underlying data. This effort resulted in the ability to use a common logic for storing and display of data on study participants, their clinical visits, and epidemiological information on their living conditions (dwelling) and geographic location. Users of the Plasmodium Genomics Resource, PlasmoDB, accessing the ICEMR data will be able to search for participants based on environmental factors such as type of dwelling, location or mosquito biting rate, characteristics such as age at enrollment, relevant genotypes or gender and visit data such as laboratory findings, diagnoses, malaria medications, symptoms, and other factors.
The ICEMR program is a global network of 10 independent
research centers created to improve understanding of the
epidemiology and transmission patterns of malaria in different
geographic regions [
Ontologies play a crucial role in heterogeneous data
integration by supporting consistent data representation and
providing a semantic framework to reveal the relationships
between data thereby facilitating information retrieval and new
knowledge discovery [
Multiple ICEMR projects have provided data for inclusion in PlasmoDB. Each ICEMR project has provided data dictionaries covering all data variables and values required for interpreting the associated data. By data dictionary, we mean a list of terms with definitions and specification of data variables, data types, format of data, and allowed values (including controlled vocabulary values). Data dictionaries are used in data exchanges among ICEMR projects and sharing with different repositories. However, data dictionaries from the different ICEMR projects generally look very different from each other in terms of type and quantity of content.
To standardize the data dictionaries from different ICEMR
projects, the variables and controlled vocabulary values were
mapped to OBO ontologies. These included the Ontology for
Biomedical Investigations (OBI) [
Supported in part by National Institute of Allergy and Infectious Diseases
National Institutes of Health, Department of Health and Human Services
Contract No. HHSN272201400030C, U.S. Public Health Service cooperative
agreements U19AI089674 (MGD) and U19AI089681 (JMV).
annotator, the BioPortal search web services [
Consistent representation of ICEMR data was achieved once the variables and values in the different ICEMR data dictionaries were either mapped to existing ontology terms or new ontology terms were created for that purpose. New ontology terms were created using two approaches.
a) If the terms were general and in a domain which have been covered by an OBO ontology, they were submitted to the relevant ontology via its issue tracker to be added in by the ontology developers. For example, disease terms were submitted to the DO tracker and terms related to the environment were submitted to the EnVO tracker.
they were added in the Eupath ontology. The Eupath ontology is an application ontology developed for providing terms to annotate data in the EuPath BRC. The EuPath ontology was built based on OBI with integration of other OBO ontologies such as PATO, OGMS, DO, etc. when needed.
The ontological mapping of data dictionary variables provides semantic clarity of types. However, organization according to term types (e.g., processes, material entities, qualities, etc.) does not necessarily provide intuitive listing on web sites for mining the data. As illustrated in Figure 1, the five main types of interest are ‘participants’, ‘dwellings’, (clinical) ‘visits’, ‘entomological measurements’ and ‘geographic location’. Therefore, we organized the data dictionary variables into categories based on their relation to these types. Within each category, the data dictionary variables are grouped based on the mapped OBO ontology terms. For example, ‘height’, ‘weight’, and ‘temperature’ (measurement data) are grouped together in the ‘physical examination’ category (which in turn is placed in the ‘visit’ category). The outcome of categorization of the variables from the multiple ICEMR data dictionaries is the ICEMR terminology and is the basis for displaying search parameters of this data on the PlasmoDB website. The ICEMR terminology is represented in the OWL format containing only ‘is a’ relations enabling visualization of the ICEMR data dictionary hierarchy organization using ontology editors. WebProtege [5] is a webbased collaborative ontology development platform and provides a means for domain experts to review and post comments on terms. We uploaded the ICEMR terminology to WebProtege and used it for collaboratively reviewing both the organization of the ICEMR terminology and the labels of terms to be displayed on the PlasmoDB web site before loading the ICEMR data into the database. This approach ensured that the data was correctly displayed on PlasmoDB for each ICEMR project. For the ICEMR terminology, we specified display labels using the rdfs:label annotation property as they are the default term labels rendered on WebProtege. In addition, we used annotation properties to specify ontological names, definitions, whether the term was an organizing category or a variable. If the term corresponded to a data dictionary variable, then annotation properties were also used for the original variable name in the data dictionary and source, the mapped ontology term, and the ontological definition. The common display labels in the ICEMR terminology were agreed upon by the contributing ICEMR projects. Each contributing ICEMR project had variables unique to that project. Therefore, the application of the ICEMR terminology for organization of each ICEMR data dictionary resulted in different but still consistent outputs. The application of the ICEMR terminologies to the different projects can be viewed at the WebProtege site (http://webprotege.stanford.edu/) as “ICEMR Amazonia”, “ICEMR Indian”, and “ICEMR PRISM” (Uganda ICEMR project).
Longitudinal data from three ICEMR projects with studies in Uganda, India, and Amazonia were submitted for inclusion in PlasmoDB. Data and data dictionaries from the Uganda and Indian ICEMR projects were provided in English whereas data and the data dictionary from the Amazonia ICEMR project were in Spanish. The Amazonia ICEMR project also provided a translated data dictionary in English. All three ICEMR projects provided participant data, dwelling data on participants, and participant-associated clinical visit data. The Uganda ICEMR project also submitted entomological measurement data.
The Amazonia ICEMR data dictionary included 84 variables and 179 controlled values for 26 variables. The Indian ICEMR data dictionary contained 118 variables with 149 controlled values for 32 variables. The Uganda ICEMR data dictionary contained 121 different kinds of variables and 481 controlled values for 21 variables.
Variables and values specified in the ICEMR data dictionaries were mapped to 10 different OBO Foundry ontologies (listed in the Methods). Table 1 lists the mapping results for each ICEMR project. A total of 209 new terms were added to the EuPath ontology for unmapped ICEMR variables. The EuPath ontology can be viewed on the WebProtege site (http://webprotege.stanford.edu/) as the “EuPath ontology” project.
Data dictionary variables from the different ICEMR projects referring to the same thing were often different. For example, “edad” in the Amazonia ICEMR data dictionary, “age_en” in the Indian ICEMR data dictionary, and “age” in the Uganda ICEMR data dictionary all refer to participant age at the time of enrollment and mapped to the ontology term EUPATH_0000120: ‘age since birth at time of enrollment’. As another example of the encountered heterogeneity, Table 2 shows a sampling of mapping between symptom related variables to ontology terms.
Ontology term mapping was also performed on the controlled values of variables. 413 controlled values used in the Uganda ICEMR data were mapped to OBO ontology terms. The remaining 68 unmapped terms were added into the EuPath ontology. Few corresponding ontology terms were found for the controlled values in the Amazonia and Indian ICEMR data (14 for Amazonia and 5 for Indian, respectively). For those values without mapped ontology terms, we have created standardized labels and will add the terms to either OBO ontologies or EuPath ontology as described in the Methods.
After ontology term mapping and standardization of value labels across data from multiple ICEMR projects, we generated (data dictionary to standardized) term mapping files for each ICEMR. These mapping files were used in the ICEMR project data loading process and enabled consistent data representation in the PlasmoDB database. dDiacttaionary Ontology term ID Ontology term label InCamEMeR display abdominalpain HP_0002027 Abdominal pain Abdominal pain apainduration EUPATH_0000154 duration of abdominal Abdominal pain pain duration Anorexia SYMP_0000523 anorexia Anorexia aduration EUPATH_0000155 duration of anorexia Anorexia duration Cough SYMP_0000614 cough Cough cduration EUPATH_0000156 duration of cough Cough duration Diarrhea DOID_13250 diarrhea Diarrhea dduration EUPATH_0000157 duration of diarrhea Diarrhea duration Fatigue SYMP_0019177 fatigue Fatigue fmduration EUPATH_0000158 duration of fatigue Fatigue duration febrile EUPATH_0000097 febrile Febrile fever EUPATH_0000100 subjective fever Fever (subjective) Headache HP_0002315 Headache Headache hduration EUPATH_0000159 duration of headache Headache duration Jaundice HP_0000952 Jaundice Jaundice jduration EUPATH_0000160 duration of jaundice Jaundice duration jointpains SYMP_0000064 joint pain Joint pains djointpains EUPATH_0000161 duration of joint pains Joint pains duration muscleaches EUPATH_0000252 Muscle aches Muscle aches mduration EUPATH_0000162 duration of muscle Muscle aches aches duration rfa OGMS_0000015 clinical history Other medical complaint seizure SYMP_0000124 seizure Seizures sduration EUPATH_0000163 duration of seizures Seizures duration fduration EUPATH_0000164 duration of subjective Subjective fever fever duration Vomiting HP_0002013 Vomiting Vomiting vduration EUPATH_0000165 duration of vomiting Vomiting duration
For each ICEMR project, around 100 different variables can be used to search and retrieve the data. As indicated in the Introduction, malaria researchers are interested in mining the data for insights about the connections between study participants, their living conditions (dwelling), their health status (clinical visit), their geographic location and exposure to mosquitos (entomological measurement data). We assigned the variables to these five categories based on their mapped ontology terms taking into account whether they were a subclass of or having a logical connection to the categories. With the exception of geographic location, each category had around 20 different variables that required further grouping to provide intuitive access to the data for end users. Further grouping was made based on the ontological understanding of data. For example, height, weight, and temperature data are all generated by physical examination. Thus, a new class of data OGMS_0000083: ‘physical examination’ was added under category ‘visit’. Using this approach, around 5 different subtypes were created under each category (except ‘geographic location’). For example, in addition to ‘physical examination’, ‘medication’, ‘diagnosis’, ‘symptoms’, ‘laboratory findings’, ‘visit type’ and ‘visit details’ were added as subtypes of the category ‘visit’.
Term labels used in an ontology are typically chosen for ontological clarity and can be quite long. As a result, such labels are often not user-friendly or practical for providing searches on web sites like PlasmoDB. Alternative display names were therefore generated for ontology terms. For example, the display name ‘Age at time of enrollment’ is used for ontology term EUPATH_0000120: ‘age since birth at time of enrollment’.
Figure 2 shows the organization of variables that will be
displayed on the website in the three ICEMR projects
discussed here using Protégé, an OWL editor [
Related but different semantic approaches were used to address the dual challenges of standardizing data dictionaries across projects and generating user-friendly displays to search and explore the associated data.
Our approach for standardization is to relate all variables
and associated values to terms from interoperable ontologies
listed at the OBO Foundry. OBO Foundry ontologies provide
the benefit of wide coverage but can also be selectively
imported to create an application ontology such as the EuPath
ontology. When existing terms were not available for mapping,
new ones were created for introduction into the source
ontologies or just placed in the application ontology. The use
of the Basic Formal Ontology (BFO) [
An ICEMR terminology was created for the purpose of web display to organize the standardized variables according to ways that users are expected to browse them. The ICEMR terminology also takes into account the need for shortened names on a web form. Underlying all the terms however is their basis for understanding through mapping to OBO / EuPath ontology terms.
The separation of web display and variable standardization provides for flexibility in providing different emphases in data browsing while maintaining the same underlying semantics. The overall approach has allowed us to achieve the goal of providing a common system with consistent representation for the three currently participating ICEMR projects. It also provides a flexible existing system for introducing data from other ICEMR projects or other studies of the same type.
We acknowledge the developers of the Disease Ontology, the Environmental Ontology and the Drug Ontology for adding our requested terms into their respective ontologies. G.C.E and D.G thank Carmen Puemape and Mitchell Guzman for excellent technical assistance in data management. [7] The Phenotype And Trait Ontology (PATO) [online]. Available: https://github.com/pato-ontology/pato/ [8] The Ontology for General Medical Sciences (OGMS) [online].
Available: https://github.com/OGMS/ogms/ [9] P. L. Buttigieg, N. Morrison, B. Smith, C. J. Mungall, and S. E. Lewis, “The environment ontology: contextualising biological and biomedical entities,” J. Biomed. Sem. vol. 4, pp. 43, December 2013.