A repository of clinically associated Staphylococcus aureus (Sa) isolates is used to semi‐automatically generate a set of application ontologies for specific subfamilies of Sa‐related disease. Each such application ontology is compatible with the Infectious Disease Ontology (IDO) and uses resources from the Open Biomedical Ontology (OBO) Foundry. The set of application ontologies forms a lattice structure beneath the IDO‐Core and IDO‐extension reference ontologies. We show how this lattice can be used to define a strategy for the construction of a new taxonomy of infectious disease incorporating genetic, molecular, and clinical data. We also outline how faceted browsing and query of annotated data is supported using a lattice application ontology.
One of the more ambitious goals of current clinical and biomedical research is the personalization of medicine, in which treatments are selected on the basis of patient-specific as well as disease-specific information. Recent advances in high-throughput technologies have resulted in a push for the use of patient-specific information in care decisions, particularly genomic and functional genomic data, but also proteomic, metabolomic, and cytometry data. It is widely believed that the increased precision of personalized medicine will yield more effective treatments, with better outcomes and fewer adverse side effects.
Personalized medicine requires that genomic (and other) data be effectively classified and associated with known clinical phenotypes and disease types. Currently available taxonomies of disease do not support this, however, and are in general not well suited for integration and analysis of high-throughput molecular and cellular data with clinical data, such as the data found in electronic medical records. Current disease taxonomies were developed primarily to support diagnosis and reimbursement coding rather than as biological representations of disease. As a consequence, they are based on single, rigid hierarchies that do not reflect the complex interconnections between disease types; they lack links to molecular- and cellular-level data and information; and they lack the sort of formal structure that would support their use for the kinds of computational analyses applied in biological and clinical research. For example, the International Classification of Disease (ICD) version 9 includes catch-all codes such as “[041.19] Other Staphylococcus” and scattered exclusions such as “[041] Bacterial infection in conditions classified elsewhere and of unspecified site. Excludes: septicemia (038.0 – 038.9)”.
The National Academies of Science have recently called
for a new taxonomy of disease, along with informatics tools
to support its construction
We believe that biomedical ontologies will be essential to
the construction of the envisioned taxonomy of disease,
especially the ontologies in the Open Biomedical Ontology
(OBO) Foundry
The Infectious Disease Ontology (IDO) suite of ontologies is being developed within the OBO Foundry framework and includes a hub – the IDO-Core – consisting of terms and relations relevant to infectious diseases generally, together with a set of disease-specific extensions derived therefrom. The IDO ontologies are interoperable and jointly cover the infectious disease domain. Here we illustrate how the IDO ontologies can be used in the construction of a part of the new taxonomy of disease and to integrate clinically relevant phenotypic and genotypic data.
We take as our case study infectious diseases caused by Staphylococcus aureus (Sa) infection. We show how isolate data from the Network on Antimicrobial Resistance in Staphylococcus aureus (NARSA) can be annotated using IDO and its extensions. We then demonstrate a faceted browser in which both phenotypic and genotypic aspects of the IDO-annotated isolate data can be exposed and queried. Our goal is to provide a resource from which an IDOconformant application ontology can be derived for a specific Sa infectious disease type. Such application ontologies can be generated in a semi-automated way and collectively form a lattice structure beneath IDO-Core (described below). While our example narrowly focuses on properties of infectious agents, this effort is part of a larger effort to create an ontological representation of Sa diseases, and we believe the same approach can be applied to host data and to the integration of host and pathogen data. 2
IDO-Core includes terms relevant for infectious diseases generally, terms such as ‘host’, ‘infectious agent’, ‘fomite’, and ‘virulence factor’, and the relations between the corresponding types. Disease- and pathogen-specific extensions are developed by extending the core to include terms and relations relevant to the corresponding infectious disease(s). For example, the IDO extension for Sa (IDO-Sa) includes terms such as ‘Staphylococcus aureus bacteremia’ and ‘Staphylococcal cassette chromosome mec’.
IDO extensions are currently being developed for influenza, malaria, brucellosis, HIV, and Sa. Further extensions will involve the creation of specific application ontologies by IDO user groups. It will be necessary for these ontologies to import terms from several OBO Foundry ontologies, as well as from existing IDO extension ontologies. This will give rise to a lattice structure beneath IDO core and its extensions, as illustrated in Figure 1. At the bottom of the lattice is IDO-ALL, the (pre-inference) closure of possible the IDO ontologies.
When a new application ontology is needed, its position in the lattice will be determined by the terms it needs to import. IDO Core is agnostic to biological scale, host organism, and disciplinary perspective, but it will be desirable for some of the application ontologies in the lattice to hold some of these fixed (e.g., genetic aspects of influenza in birds), thus serving as granular partitions of the domain ontology they are extending. The lattice serves as a representation of some of the interdependencies in the existing IDO set of ontologies and the intended overall domain coverage.
Fig 1. A possible lattice expansion of IDO 2.1
The IDO ontologies represent disease according to the disorder – disease – disease course framework provided by the Ontology for General Medical Science (OGMS), in which a disorder is the physical basis of a disease, which is itself a disposition to pathological processes realized in a disease course. For example, in IDO-Sa we assert the following in OWL-DL: Sa subClassOf obi:organism AND
ido:‘infectious agent’ SaI =def ido:‘infectious disorder’ AND
has_part SOME Sa SaID =def ido:‘infectious disease’ AND
has_material_basis_in SOME SaI.
SaID realized_by ONLY SaIDC where, ‘Staphylococcus aureus’ = Sa, ‘Sa Infectious Disorder’ = SaI,‘Sa Infectious Disease’ = SaID, and ‘Sa Infectious Disease Course’=SaIDC.
The primary classification of Sa is as an organism, but Sa bacteria are also infectious agents because they have a disposition to cause infectious disease in some hosts. Note we define Sa infectious disorder as an infectious disorder that has Sa as part, but we do not assert “Sa part_of SOME SaI” because Sa can be among a host’s normal flora, for example on the skin or nasal mucosa.
We use the shortcut relation has_material_basis here to establish a link between the disease (disposition) and the disorder (material entity) (Goldfain, Smith and Cowell, under review). An infectious disorder is both an infection (a material entity composed of infectious agents) and a disorder (has reached the threshold of clinical significance to dispose a host to infectious disease). 2.2
Classifying Staphylococcus aureus diseases Infectious diseases can usefully be classified in terms of a number of differentia, including: host type, (sub-)species of infectious agent, route of transmission, antibiotic resistance, and anatomical site of infection.
For many species of infectious agent, including Sa, a further classification into strain categories is useful. Many different typing systems are used, including: Pulse Field Gel Electrophoresis (into strains), Multi-Locus Sequence Typing (into sequence types), BURST Clustering (into clonal complexes), and gram staining (into gram positive and gram negative classes). Each of these typing systems is tied to a particular type of assay that can be described using the Ontology for Biomedical Investigations (OBI).
For our present purpose, we are interested in a typing
system specifically created to differentiate Sa isolates, the
Staphylococcal cassette chromosome mec (SCCmec) typing
system. SCCmec is further differentiated by its subparts: (a)
Cassette chromosome recombinases (ccr) and (b) mec gene
complex (mec). The SCCmec is a mobile genetic element
that carries the central determinant for broad-spectrum
betalactam antibiotic resistance encoded by the mecA gene
SCCMec subClassOf so:mobile_genetic_element ‘mec gene complex’ subClassOf
so:gene_cassette_member ‘ccr gene complex’ subClassOf
so:gene_cassette_member SCCMec has_part SOME ‘mec gene complex’
SCCMec has_part SOME ‘ccr gene complex’ The classification of SCCmec as a gene cassette is to be preferred over its classification as a mobile genetic element because the former tells us what SCCmec is, while the latter tells us what SCCmec can do. However, we include both here, because most descriptions of SCCmec highlight its mobility. Description of a SCCMec subtype then proceeds as follows:
1 http://www.sccmec.org/Pages/SCC_ClassificationEN.html
3
We will now show how a lattice of Sa isolates can be constructed using IDO-Sa and isolate metadata indicating properties such as the mec and ccr gene complex types. The isolate lattice is then used as the basis for our desired lattice of infectious disease application ontologies. Ontologically speaking, isolates are particulars that instantiate the organism type Sa and have been extracted from a host organism. Here we do not represented the distinctions between Sa as an ‘isolate’ or as part of a ‘cell culture’, however we believe these terms are general enough to infectious disease research to warrant inclusion in IDO-Core.
The ontology generated for this case study is stored across several OWL files. The full ontology, including external imports and automatically generated isolate information is currently available in OWL-DL format at http://www.awqbi.com/LATTICE/narsa-complete.owl. The ontology was developed using Protege 4.1 and was checked for inconsistency using the Hermit 1.3.5 and Fact++ reasoners. 3.1
Wherever possible, we import and reuse terms (and URIs)
from OBO Foundry ontologies via the MIREOT technique
We also import drug file names from the National Drug File Reference Terminology (NDF-RT) to represent antibiotic resistance, and create links to two other resources: (1) Antibiotic Resistance Ontology8 and Antibiotic Resistance Database Ontology9. Various other stakeholders (such as the DebugIT European Union initiative) have ontologies and databases of antimicrobial resistance, but we only to link to open resources for our case study. 2 http://code.google.com/p/ogms/ 3 http://obi-ontology.org/page/Main_Page 4 http://infectiousdiseaseontology.org/page/Main_Page 5 http://code.google.com/p/information-artifact-ontology/ 6 http://www.ncbi.nlm.nih.gov/Taxonomy/ 7 http://sig.biostr.washington.edu/projects/fm/ 8 http://arpcard.mcmaster.ca 9 http://ardb.cbcb.umd.edu/antibio_resis.obo 3.2
The Network on Antimicrobial Resistance in Staphylococcus aureus10 maintains a repository of Sa isolates for clinical research which includes genetic, phenotypic, and demographic information on each isolate. For this example, we use a subset of 101 NARSA isolates, those listed in the “Known Clinically Associated Strains – ABCs Collection from CDC” repository. All of the isolates in this subset have an SCCMec type annotation in the NARSA repository and have diverse geographic origin in the United States.11
The NARSA subset was selected to demonstrate how a disease lattice could be constructed starting from only structured HTML content about isolates. NARSA maintains a database of extended information about such isolates; however we only used the information publicly available on the web.
A script was created to extract each isolate’s NARSA id (NRSnnn), culture source, toxin profile, and antimicrobial profile. The script was implemented in Ruby and utilized the Hpricot HTML library and regular expressions to extract information. First, the NARSA id was used to assert the existence of a Sa instance type. Then the culture source data was extracted. The culture source was sometimes unspecified (‘other’) or underspecified (‘blood’ vs ‘wound’). Only culture sources for which FMA types existed were asserted to exist as such, but IDO allows for an even more complete representation of host anatomical entities if such information is known. For example, the anatomical location from which the infectious organism is isolated may also be a portal of entry.
The toxin profile for NARSA subset isolates included the presence or absence of the Panton Valentine Leukocidin (PVL) and Toxic Shock Syndrome Toxin (TSST). These toxins are strong determinants of the virulence and clinical manifestation of Sa disease. We classify PVL and TSST as ido:exotoxin. The presence or absence of a toxin is not usually associated with drug resistance, but by representing both pieces of information we are able to query the application ontology for correlations between the presence of toxins and resistance to certain drug types.
The antimicrobial profile for the NARSA subset includes 15 drugs (see Figure 2 for a subset of these). For each drug, NARSA reports a minimum inhibitory concentration – a range or exact value – along with an interpretation of the antibiotic resistance indicated by this value following the Clinical and Laboratory Standards Institute guidelines. 10 See http://www.narsa.net/ 11 See http://www.cdc.gov/abcs/reports-findings/surv-reports.html
Fig 2. Antimicrobial profile for an isolate in the NARSA subset The NDF-RT was used to validate this profile by making sure that the set of drugs in the profile is a subset of: {d | ndf-rt:’Staph Infection’ ndf-rt:may_be_treated_by d} For NARSA, or any other resource on antimicrobial resistance, there may be a good reason to restrict attention to a subset of antimicrobials. However, since new resistance evolves rapidly, a resource such as NDF-RT can be used to synchronize the latest antibiotics permissible in such a profile.
Minimum inhibitory concentration data (MIC) are represented using IAO and OBI as follows: ‘MIC assay’ subclassOf iao:assay ‘MIC assay’ has_specified_output SOME
‘MIC data item’
‘MIC scalar measurement datum’ is_about SOME
‘drug susceptibility of infectious agent’
Resistance is a disposition that an infectious agent bears
towards some drugs and is realized in their presence. We
have elsewhere modeled resistance in terms of pairwise
complementary dispositions on the part of both the
infectious agent and the drug
measurement datum’)
Finally, for each drug D towards which the isolate Sa has a drug resistance we assert: ‘resistance to D’ subclassOf
ido:‘resistance to drug’
Sa has_disposition SOME ‘resistance to D’ 3.3
The lattice of infectious diseases mirrors the isolate lattice by representing the types of infectious disease different isolates can give rise to. Infectious agents are parts of those infectious disorders which are the material basis for infectious disease. Using the representation developed above, we can begin to make assertions about the specific types of disease the isolates give rise to and the profiles of the disease courses which realize these diseases. For example, the presence of the PVL toxin in Sa can lead to necrotic lesions (ogms:disorder) and necrotizing pneumonia (ogms:disease).
A faceted browser of the ontologically annotated NARSA isolates was constructed using the MIT Exhibit 2.0 library (http://www.awqbi.com/LATTICE/narsa-complete.html). This tool allows the user to visualize and correlate isolate information across different dimensions (see Figure 3).
Fig 3. Faceted browsing illustrates that most isolates with a resistance to Clindamycin are of SCCmec type II and lack PVL Linking to external resources is facilitated by the fact that such facets are assigned ontology types from the IDO lattice. These are exactly the kinds of links that will be needed for the knowledge network supporting a new taxonomy of disease.
A lattice of infectious disease ontologies can serve as a mechanism to integrate pathogen-specific typing systems such as SCCMec with phenotypic data such as drug resistance. Such genotype-phenotype relations will be the key to a more effective taxonomy of disease that enables truly personalized medicine. The lattice of infectious diseases is expected to grow along predictable dimensions (host organism, infectious agent organism, drug resistance), but can accommodate lightweight application ontologies that are created for very specific purposes. Each such application ontology will have a place in the lattice on the basis of what IDO terms it imports.
We have shown that IDO-conformant annotation of isolate data (such as that in the NARSA repository) is possible without the need to reassemble OBO Foundry resources for new applications. Other benefits of our approach include: exposing currently accepted SCCmec types in a computable format via an ontology and validating the NARSA antimicrobial profile using the NDF-RT.
We hope to reuse a similar technique to that outlined in this paper for isolate repositories across the infectious disease domain. In so doing, we hope to broaden the lattice and integrating organism specific typing systems with the IDO suite of ontologies. We believe that such an effort can be a powerful enabler for a new taxonomy of infectious disease and its supporting knowledge network.
This work was funded by the National Institutes of Health through Grant R01 AI 77706-01. Smith’s contributions were funded through the NIH Roadmap for Medical Research, Grant U54 HG004028 (National Center for Biomedical Ontology).