-The Apollo Structured Vocabulary (Apollo-SV) is a Web Ontology Language 2 (OWL 2) representation of terms related to epidemic simulation. We are developing Apollo-SV by ontological analysis of the information used and created by epidemic simulators and the entities this information is about. A key finding of our analysis is that the input of an epidemic simulator is properly understood as (1) a representation of an ecosystem at simulator time zero, (2) information about infectious diseases of interest in the ecosystem, and (3) information about plans to control the diseases. This insight is reflected in the scope of Apollo-SV, which includes terms from the domains of both infectious disease epidemiology and population biology. We also found that some definitions in the Infectious Disease Ontology (IDO), including 'infection', 'infection acquisition', 'infectious disease', 'pathogen', and 'host', were not compatible with the meanings of the terms as used in epidemic simulation; thus, we created new definitions of these terms. Our analysis of epidemic simulators-which are mathematical models of phenomena studied by infectious disease epidemiology-afforded several advantages that likely explain why we discovered limitations of IDO. As a result, we recommend that development of biomedical ontologies intended for reuse consider the perspective of the overlapping biological science(s) involved. Apollo-SV is freely available at: http://purl.obolibrary.org/obo/apollo_sv.owl.
The science and practice of infectious disease epidemiology, like climate science, is increasingly reliant on computational simulation. The simulators—known as epidemic simulators or more generally disease transmission models (DTMs)1—require machine-interpretable information about pathogens, rates of
In this paper, we describe one element of our proposed
standards—the Apollo Structured Vocabulary (Apollo-SV).
Apollo-SV is an OWL 2 representation of terms related to
epidemic simulation. The other two elements are an XML
Schema Document (XSD), which defines the syntax for
simulator input, and a database schema that defines the
representation of simulator output. Apollo-SV defines the
terminology used in the XSD and database schema. These
elements are described in Wagner et al. [
METHODS We developed Apollo-SV for use in a set of Web services designed to improve access to epidemic simulators. We begin this section with an overview of these services, then detail our methods to develop Apollo-SV, including conformance to OBO Foundry Principles, methods to ensure validity for its intended use, and multi-disciplinary development.
The Apollo Web Services: Briefly, the Apollo Web Services are a set of Web services designed to allow a publicly available, Web-based, end-user application to access multiple epidemic simulators through requests to a single Broker service (Fig. 1).
In Figure 1, the Simple End User Application (SEUA) creates an infectious disease scenario for simulation, encoded in an XML document that conforms to the Apollo XSD, which uses terminology defined by Apollo-SV. The SEUA invokes the runSimulation() method of the Broker service with the infectious disease scenario. The Broker service invokes the Translator service, which translates the infectious disease scenario into the native terminology and syntax of the requested simulator(s).
In accordance with Foundry principles, we write a textual definition for every term that we create. Because formal ontological textual definitions often use the technical language of ontologists, we created an elucidation annotation for classes in Apollo-SV. The elucidation restates the definition in language more familiar to subject matter experts. We also axiomatize Apollo-SV terms wherever possible (e.g., Fig. 2-5).
In accordance with the Foundry principle of orthogonality,
which stipulates that a given term is defined only once across
all ontologies, we search for and import pre-existing
ontological representations into Apollo-SV. Besides importing
entire ontologies, we import selected classes, individuals, and
properties using the Minimum Information to Reference an
External Ontology Term (MIREOT) [
We also adhere to Foundry naming conventions [
To help link the OWL file with the XSD, we create a Unique Apollo Label (UAL) for classes in Apollo-SV. The UAL is the exact XSD type or attribute name to which the class in Apollo-SV corresponds, for example, InfectiousDisease and basicReproductionNumber.
Analysis of simulators’ input and output files, and documentation: We analyzed the input and output files of four epidemic simulators. We also analyzed documentation, such as user guides and published papers. We reviewed terms that we extracted from these resources with the developers of the simulators to identify relevant but missing terms, to discover synonymy among terms, and to detect and resolve ambiguity.
further refine our OWL DL representations by using the terms in the XSD representation as it progressively expands to be able to represent the input of four simulators.
developing the mappings from the XSD and Apollo-SV terms to the native language of the simulators identifies additional issues with Apollo-SV that we feed back into our analysis.
exposes Apollo-SV definitions and elucidations in tool tips that appear when the mouse hovers over a term. This view identifies problems with elucidations by placing them into the context of an end-user configuring a simulator, and wanting to understand what is meant by a term.
Public release: To encourage adoption of Apollo-SV and to allow external scientific review, comments, and requests for additions, we make Apollo-SV publicly available at http://purl.obolibrary.org/obo/apollo_sv.owl, a permanent URL (PURL). We also ensure that Apollo-SV is easily accessible for browsing and download at the Web-based Ontobee portal: http://www.ontobee.org/browser/index.php?o=APOLLO_SV. The issue tracker and under-development version of ApolloSV are located at our Google Code site. The PURL to the development version of Apollo-SV is http://purl.obolibrary.org/obo/apollo_sv/dev/apollo_sv.owl.
SV comprises personnel with backgrounds in simulator development, disease surveillance, medicine, biomedical informatics, medical terminologies, ontological engineering, artificial intelligence, and formal logic. All these individuals, including a simulator developer (author SB), have been actively engaged in review of Apollo-SV, and their feedback guides design decisions.
III.
RESULTS
Overall, Apollo-SV has 594 classes: 287 that we created new in Apollo-SV and 307 that we imported: 57 via MIREOT (Table I) and 250 from entire ontologies. The number of imported classes is artificially high because the import of entire ontologies brings classes into Apollo-SV we do not require.
The core classes in Apollo-SV represent key entities of interest to infectious disease epidemiology and population biology (Table II). Throughout the course of developing the Apollo standard, we reached the conclusion that the input to an epidemic simulator is properly understood as a representation of an ecosystem at simulator time zero, with additional information about infectious diseases and planned or ongoing interventions to control them. This conclusion motivates the inclusion in Apollo-SV of terms from population biology. In turn, the ecosystem viewpoint heavily influenced our definitions of key terms in infectious disease epidemiology.
At present, Apollo-SV and the XSD enable configuration of three epidemic simulators with the same infectious disease scenario in the SEUA. We are piloting a fourth simulator. They are (1) a compartmental model developed by authors MMW, NEM, and JDL (disease agnostic); (2) the FRED model developed by the University of Pittsburgh Public Health Dynamics Laboratory in collaboration with the Pittsburgh Supercomputing Center and the School of Computer Science at Carnegie Mellon University, University of Pittsburgh and Imperial College (influenza A in humans); and (3) the FluTE model developed by the University of Washington and Fred Hutchinson Cancer Research Center in Seattle and the Los Alamos National Laboratories (influenza A in humans).
With respect to Foundry orthogonality, we attempted to reuse IDO’s definitions of ‘infection’, ‘pathogen’, ‘host’, but had to create new definitions (and thus new representations) for them in Apollo-SV as discussed in the following sections.
TABLE II. CLASSES IN APOLLO-SV BY SUBDOMAIN Domain Infectious disease epidemiology Population biology
Classes in Apollo-SV Infection Pathogen Latent period Contaminated thing Contamination Infectious disease scenario Transmission coefficient Disease transmission model Susceptible population Infectious population Ecosystem Abiotic ecosystem Population Population infection and immunity census
Infection acquisition Host Infectious period Contamination acquisition Basic reproduction number Transmission probability Infectious disease control strategy Exposed population Resistant population Biotic ecosystem Community Population census Abiotic ecosystem census abnormality to immunocompetent organisms of the same
However, epidemic simulators represent infection as a
process, because that is how ‘infection’ is defined in infectious
disease epidemiology. For example, [
Also, the IDO definition requires that an infection cause clinical abnormality in an individual of a particular species. However, infectious disease epidemiology recognizes the existence of species that do not experience clinical abnormalities when infected with a particular pathogen. The importance in epidemic simulation is that members of species that can experience clinical abnormalities when infected can acquire infection with the pathogen from a ‘carrier’ species.
Apollo-SV defines ‘infection’ as: A reproduction of a pathogen in (a part of) the tissue of an organism from another species (Fig. 2).
This biologically-grounded definition recognizes that two species are interacting and—from the pathogen species point of view—infection is a process of reproduction. The definition only requires reproduction of one species within the tissues of an individual (organism) from another species.
IDO imports two definitions of ‘transmission process’ from the Transmission Ontology:
transmitted directly or indirectly from its natural reservoir, a susceptible host or source to a new host.
The second, “suggested definition” erroneously restricts transmission to occur only between two hosts of the same species. It is thus not usable in infectious disease epidemiology or any other science that deals with cross-species transmission.
The first definition of ‘transmission process’ has two major problems. The first problem is that it is circular, defining ‘transmission process’ in terms of a pathogen being transmitted, with no definition of ‘transmitted.’
The second problem is an ontological one. It attributes to one process the property of being the means by which something else happens. For example, assume droplet spread of infection from one host to another by a sneeze. This definition equates the sneeze with the transmission process. That is, it says that only the sneeze exists, but it also has the property of “having transmitted the pathogen”. However, equating the sneeze to the transmission process is nonsensical because for transmission to be complete, the second host must have an infection. But this infection will not begin for minutes to hours after the sneeze is over. The sneeze cannot somehow extend itself in time until an infection is established, but conversely not extend in time when no infection results. There exist two distinct processes: the sneeze and the transmission.
We also had the insight that it is only the second host who acquires the infection that undergoes a change during the process. Therefore, we chose to rename it ‘infection acquisition’. We recognize that we are diverging from standard terminology in the field, but anyone wishing to add an alternative label to the infection acquisition class in Apollo-SV could do so without changing the meaning of the class.
Apollo-SV defines ‘infection acquisition’ as: The biological process of pathogen organism(s) entering (the body of) a host organism from a contagious host or a contaminated thing and reproducing using host resources.
As with our definition of ‘infection’, this definition is biologically grounded and recognizes that from the pathogen species’ point of view, infection acquisition is the entry into a host and the beginning of reproduction there. Note that ApolloSV’s definition of ‘contaminated thing’ is general and includes natural reservoirs, vector organisms that are not infected (a.k.a. mechanical vectors), and fomites like contaminated pencils.
This definition is not sufficient in and of itself to understand what IDO refers to by ‘host’. It is also necessary to review its definitions of ‘host role’ and ‘extended organism’: 1. ‘Host role’: A role borne by an organism in virtue of the fact that its extended organism contains a material entity other than the organism.
Under these definitions, any organism that has an artificial joint, a penny in its gut, or an arrow through its chest is a host. The fact that a person with a prosthetic knee is a “host” is counterintuitive. This definition is too admissive for our use cases (and for clinical medicine, too): any foreign material entity inside the organism’s body renders the organism a host.
In addition, from the ontological perspective, we doubt there is any such entity as host role. First, according to BFO, a role is manifested or realized in one or more processes. However, because there is no representation of the infection process in IDO, infection cannot be the realization. No other process in IDO suffices, either. If there is no process that realizes a role, then by definition of ‘role’, there is no role.
We therefore distinguish pathogen and host based on which one is reproducing inside tissue (pathogen) and which one is the location of the reproduction (host).
IDO defines ‘pathogen’ as: A material entity with a pathogenic disposition.
Again, this definition requires the definitions of other terms to understand its meaning: 1. ‘Pathogenic disposition’: A disposition to initiate processes that result in a disorder.
Thus per IDO any material that causes injury is a pathogen, including the endotoxin of Clostridium difficile or an overdose of acetaminophen. IDO does have an infectious agent class as a subtype to pathogen that refers specifically to organisms that enter into a host cause injury. But this definition is not how infectious disease epidemiology uses the term ‘pathogen’.
IDO also asserts pathogens must typically cause disease. However, attenuated poliovirus used in oral polio vaccines infects the gut mucosa of humans and thus is a pathogen (or infectious agent per IDO), but it causes disease in only one per 2.7 million first doses of vaccine.
Apollo-SV defines ‘pathogen’ as: A material entity that is the bearer of a disposition that, when realized, is realized as an infection (Fig. 4).
Per IDO, infectious disorder is a subytpe of infection. However, we require a representation of infectious disease that is consistent with our definition of ‘infection’ as a process. But because IDO defines ‘infection’ and thus by inheritance ‘infectious disorder’ as a material entity, we could not reuse this definition of ‘infectious disease’.
Apollo-SV defines ‘infectious disease’ as: A disease that
inheres in a host and, when realized, is realized as a disease
course that is causally preceded by an infection (Fig. 5). This
definition is compatible with the OBO Foundry definition of
disease, which is in the Ontology of General Medical Science
(OGMS) [
Note that the disease inheres only in the host. From the pathogen’s perspective, there is no clinical abnormality (which is a necessary condition to meet the definition of disease in OGMS). For the pathogen, infection is perfectly normal.
IV.
DISCUSSION Apollo-SV version 2.0.1 is an ontology for use in representing DTM input and output. It includes core terms from infectious disease epidemiology and population biology. Apollo-SV currently supports the representation of infectious disease scenarios that can be run on three epidemic simulators and the results of the simulations.
We set a high priority on implementing Apollo-SV in a Web service for three reasons. First, we wanted to demonstrate the capability to initialize multiple simulators with one infectious disease scenario to motivate the adoption of Apollo. In addition, we wanted to lower barriers to adoption by making available a reference implementation. Lastly, implementation is the basis of our iterative development and refinement process that ensures Apollo is production ready and flexible, which also lowers the barriers to adoption.
A key insight from our iterative development of Apollo-SV and XSD is that a simulator configuration is properly understood as a representation of an ecosystem at a particular time. This insight led us to include in Apollo-SV key terms from population biology, such as ‘ecosystem’ and ‘census’. Furthermore, it led us to our redefinition of ‘infection’, which was central to redefining other IDO terms.
Fig. 5. Representation of the equivalent class axiom for "infectious disease" in Apollo-SV. The graphical representation is analogous to Fig. 1.
A key result of our development of Apollo-SV was that we could not reuse IDO definitions of ‘infection’, ‘host’, ‘pathogen’, and ‘infectious disease’, and thus we created the definitions presented here. This result was surprising because we had anticipated reuse of IDO at the outset of Apollo-SV development. Given that we did not expect this result, it is worth considering the possible reasons behind it.
A key reason is that our concentration on how terms are used in biological sciences—especially population biology— exposed many issues. This focus differed fundamentally from IDO’s concentration on how the terms are used in clinical medicine. In particular, our focus led us to a requirement to represent the process of infection as opposed to the steadystate, material-entity view of IDO.
So what then led us to the perspective of population biology? We believe the reason we reached this perspective, as well as ontological clarity elsewhere in Apollo-SV, is that working with epidemic simulators quickly brought into view the key phenomena studied (that are also of relevance to epidemic simulation) and their fundamental nature. These simulators are mathematical models in the field of infectious disease epidemiology. They have explicit ontological commitments that have been rigorously vetted through peer review of research using the models (as well as the models themselves). In addition, these ontological commitments comprise the core entities involved in infections and their acquisition, leading us to confront the issues involved in representing them from the outset. It is likely that IDO, by contrast, started with a disease focus and worked from there towards the nature of infection. Finally, because these simulators make a relatively small number of ontological commitments, we had the ability to to devote sufficient time to ontological analysis while still achieving demonstrable progress in implementing the SEUA. Because implementing the SEUA is central to validating Apollo standards, we thus also achieved rapid validation.
We conclude that biomedical ontology developers should incorporate the perspective of the basic sciences that study the phenomena underlying clinical practice, such as medicine and public health practice, when developing ontologies of clinical phenomena. When mathematical models of such phenomena exist, they are potentially useful starting points for analysis.
We also could not reuse prior work on other ontologies that
have overlap with Apollo-SV. This work includes the
Epidemiology Ontology (EO) [
Note that Apollo-SV is still a significant work in progress. We represented entities sufficient to cover the input and output of just three simulators and did pilot work on a fourth simulator. We have also done preliminary work to represent other types of information in infectious disease epidemiology.
Our future plans include harmonizing Apollo-SV
definitions with IDO (we plan to submit the issues described
here to the IDO issue tracker) and expanding Apollo-SV to
cover additional simulators and types of information used in
infectious disease epidemiology. We also plan to study the
3 The link provided at http://surveillance.mcgill.ca/trac/star/
wiki/StarComponents/Ontology is broken as of this writing.
potential to generate the XSD from the ontology, a successful
strategy in other projects [
This work was funded by award R01GM101151 from the National Institute for General Medical Sciences (NIGMS). This paper does not represent the view of NIGMS. This work used the Protégé resource, which is supported by grant GM10331601 from NIGMS.