-Crop specific ontologies for phenotype annotations in breeding have proliferated over the last 10 years. Across-crop data interoperability involves linking those ontologies together. For this purpose, the Planteome project is mapping the Crop Ontology traits (www.cropontology.org) to the reference ontology for plant traits, Trait Ontology (TO). Manual mapping is timeconsuming and not sustainable in the long-run as ontologies keep on evolving and multiplicating. We are thus working on developing reliable automated mapping techniques to assist curators in performing semantic integration. Our study shows the benefit of the ontology matching technique based on formal definitions and shared ontology design patterns, compared to standard automatic ontology matching algorithm, such as AML (AgreementMakerLight).
The development of improved crop varieties relies on both
traditional breeding methods and next-generation methods such
as high-throughput sequencing, molecular breeding and
automated scoring of traits. In that context, a number of
ontologies have been developed to face the data
interoperability issues. They fulfill the needs of specific
communities, but are species or clade-specific ontologies [
The crop breeding community, in particular widely uses the
Crop Ontology (CO; www.cropontology.org), which is
composed of species-specific ontologies for fieldbook edition
and data annotation [
The Planteome project (www.planteome.org) aims to
support comparative plant biology, and provides integrated
access to annotated datasets generated by inter and
intraspecific comparative analysis of transcriptomes, proteomics,
phenomics and genome annotation. To address this objective,
Planteome is currently developing, and promoting the use of a
set of reference ontologies for plants, proposing species-neutral
concepts, as well as common data annotation standards.
Harmonization between the species-specific ontologies and the
Planteome reference ontologies is currently done by mapping
Crop Ontology to the Plant Trait Ontology (TO) [
The purpose our study is to generate mappings in an
efficient way in order to ease the work of the ontology curators
in creating manual mappings. In this objective, we have
compared two automatic ontology mapping techniques. The
first technique is widely used to align ontologies and consists
in applying a standard automatic matching algorithm. Indeed,
AML (AgreementMakerLight) performs mappings based on
both the string similarities of the ontology terms and the
ontology structure. Considering the number of ontologies to be
mapped and the inherent nature of ontologies to evolve over
time, it can be hard to maintain automatically the mappings
created using such a technique. Therefore, the second
technique relies on formal definitions and shared ontology
design patterns. The ontology design patterns are created using
Ontology Web Language (OWL) axioms based on
EntityQuality (EQ) statements, leading to a post-composition of
terms, similar to what has been proposed by the Ontology of
Biological Attributes (OBA) [
have been applied to four crop Trait Dictionaries of the Crop Ontology so far: cereals rice and wheat, legume lentil and root tuber crop cassava. Those ontologies are very different in terms of plant anatomy and morphology, but also in terms of count and complexity of phenotypic traits. Table 2 summarizes the results of the mappings process on trait terms. Mapping using formal definitions resulted in two-fold increase successful mappings. On average, AML was able to propose mappings for ~40% of the CO classes in each ontology compared to ~75% mapped terms using the formal definition approach. This can be explained by the fact that crop specific ontologies use very specific terminologies, especially for the Entity part of the EQ statement. Although the specific plant entities are defined in the Plant Ontology (PO) as synonyms of species neutral entities, all the synonyms were not added to TO and CO when the terms were pre-composed. The AML algorithm was thus not able to use this information, whereas the PO synonyms have been used in order to build the formal definitions of the CO terms. Furthermore, because the class hierarchy is quite simple in the different CO, AML was not able to use the ontology structures to improve the mapping results: only equivalent terms were found using AML.
Disease resistance traits are important for breeders. A disease results from the combination of a host species, a pathogen and an environment, the disease resistance traits are crop-specific. Biotic stress traits include disease-related traits and can cover as much as 20% of the individual CO. Those traits cannot have an exact correspondence in TO. Thus AML was not able to find mappings for those terms. Based on the formal definitions, a reasoner linked those terms to be subclasses of one the TO stress trait.
Finally, all the classes in TO haven’t been formally defined. Indeed, design patterns are hard to develop for very complex traits such as yield-related traits. This is why the all the CO classes cannot be mapped to TO classes using the design pattern technique. Manual mapping is still needed in order to map those traits. The mapping coverage will be improved in the future. The mapped ontologies are available on www.planteome.org as well as on Planteom’s GitHub repository (https://github.com/Planteome). # trait classes
AML Design Patterns
In an era of ontology proliferation, it is of vital importance to have reference ontologies and powerful tools that reduce the effort of ontology alignment. Standard mapping techniques do not fit the need of ontology evolution over time as their results are difficult to maintain automatically. Developing the mapping process based on ontology design patterns and logical axioms ensures validity confidence accuracy of the resulting ontology mappings. Scientists from the breeding community can continue to use the standards preferred by them to annotate/record their data, reducing the effort they need to provide. Planteome, through the TO, provides unified access to the breeding and the genetic data, opening up the possibility to perform large scale analysis such as comparative genomics by promoting a species neutral approach.
ACKNOWLEDGMENT
This work was supported by IOS:1340112 from the NSF. Additionally, CJM acknowledges the support of the Director, Office of Science, Office of Basic Energy Sciences, of the U.S. Department of Energy under Contract No. DE-AC0205CH11231