CARO: The Common Anatomy Reference Ontology


Haendel, M.A., Osumi-Sutherland, D.S. Neuhaus, F.,
Mabee, P.M., Mungall, C.J., Ashburner, M., and Smith, B.

To appear in Anatomy Ontologies for Bioinformatics: Principles and Practice
Albert Burger, Duncan Davidson and Richard Baldock, Editors
Springer, 2008, 327-349



Abstract
The Canonical Anatomy Reference Ontology (CARO) is being developed to facilitate
interoperability between existing anatomy ontologies for different species, and will
provide a template for building new anatomy ontologies. CARO has a structural axis of
classification based on the top-level nodes of the Foundational Model of Anatomy.
CARO will complement the developmental process sub-ontology of the GO Biological
Process ontology, using it to ensure the coherent treatment of developmental stages, and
to provide a common framework for the model organism communities to classify
developmental structures. Definitions for the types and relationships are being generated
by a consortium of investigators from diverse backgrounds to ensure applicability to all
organisms. CARO will support the coordination of cross-species ontologies at all levels
of anatomical granularity by cross-referencing types within the cell type ontology (CL)
and the Gene Ontology (GO) Cellular Component ontology. A complete cross-species
CARO could be utilized in other ontologies for cross-product generation.

1. Necessity of a canonical anatomy reference ontology

Genomes are modified over evolutionary time to produce a diversity of anatomical forms.
Understanding the relationship between a genome and its phenotypic outcome requires an
integrative approach that synthesizes knowledge derived from the study of biological
entities at various levels of granularity, encompassing gene structure and function,
development, phylogenetic relationships, and ecology.

Many model organism databases (MODs) collect large amounts of data on the
relationship between genetic/genomic variation and morphological phenotypes in
databases, which standardize the description of morphological phenotypes and gene
expression patterns using types from anatomy ontologies specific to their species of
interest. These ontologies have allowed the MODs to group phenotypic and gene
expression data pertaining to particular anatomical types.1 Methods of phenotype curation

1 In keeping with the nomenclature of Smith et al., 2005a, we prefer the term ‘type’ to ‘class’. Ontologies
contain terms that refer to types of things in the real world. A type should not be confused with its
instances. For example, a human anatomy ontology might contain the term ‘foot’. This refers to the type
Preprint

CARO: The Common Anatomy Reference Ontology 2


are being extended and standardized as part of the work of the National Center for
Biomedical Ontology (NCBO), which aims to provide data-mining tools which can be
applied across all species, in particular in support of queries relating to anatomical
structures and associated genes. However, there is currently no system for standardizing
the representation of anatomy in ontologies.

Cross-species standardization among anatomy ontologies would bring a number of
benefits. First, it would allow the development of standardized tools for grouping and
querying anatomy-linked data. Second, it is a prerequisite for inference of anatomically
based phenotypic and gene expression data within and across species. Third, if anatomy
ontologies were standardized, then a method for representing homology between
anatomical types in different anatomy ontologies could be devised. Fourth,
standardization would allow better interoperability between anatomy ontologies and other
ontologies.

In this chapter, we propose a canonical anatomy reference ontology (CARO), which is
designed to serve as a standardized, generic structural classification system for
anatomical entities. We also propose a standardized set of relations for use in building
anatomy ontologies, extending the set of relations already defined as part of the OBO
Relations Ontology (RO; Smith et al., 2005a). By necessity, this proposal also begins to
address the key issue of representation of homology between anatomical types in the
context of anatomy ontologies.

This chapter summarizes progress on creating CARO, drawing on conclusions reached
during an anatomy ontology workshop held in Seattle, WA, in September of 20062
sponsored by the National Center for Biomedical Ontology.

2. What is CARO?

CARO is an ontology of canonical anatomy. At its core is a single, structural
classification scheme based on that developed by the Foundational Model of Anatomy
(FMA), a well established ontology of human anatomy (Rosse and Mejino, 2003). One
reason the FMA was chosen as a model for CARO is because the FMA adheres to the
principles laid out by the OBO Foundry. CARO has adopted the policy of single
inheritance based principally on the empirical observation that ontologies that allow
multiple inheritance, while easier to build, are marked by characteristic errors, which
generally result from the use of multiple classification schemes within a single ontology,
leading to what has been called ‘is_a overloading’. This can be avoided by utilizing
genus-differentia definitions of the terms in ontologies, in which each type is specified as
a refinement (via some differentia) of an existing more general type (the genus, i.e. the
corresponding parent type, in the is_a hierarchy). Definitions of this form are typically
written along the lines of “An S is_a G which D”. This provides unambiguous definitions
that can be applied consistently and leads to clean classification hierarchies in which all

human foot, of which your left foot is an instance. The collection of all such instances is the extension of
the corresponding type.
2 http://www.bioontology.org/wiki/index.php/Anatomy_Ontology_Workshop