The Drosophila anatomy ontology

Abstract

Background: Anatomy ontologies are query-able classifications of anatomical structures. They provide a widely-used means for standardising the annotation of phenotypes and expression in both human-readable and programmatically accessible forms. They are also frequently used to group annotations in biologically meaningful ways. Accurate annotation requires clear textual definitions for terms, ideally accompanied by images. Accurate grouping and fruitful programmatic usage requires high-quality formal definitions that can be used to automate classification and check for errors. The Drosophila anatomy ontology (DAO) consists of over 8000 classes with broad coverage of Drosophila anatomy. It has been used extensively for annotation by a range of resources, but until recently it was poorly formalised and had few textual definitions.

Results: We have transformed the DAO into an ontology rich in formal and textual definitions in which the majority of classifications are automated and extensive error checking ensures quality. Here we present an overview of the content of the DAO, the patterns used in its formalisation, and the various uses it has been put to.

Conclusions: As a result of the work described here, the DAO provides a high-quality, queryable reference for the wild-type anatomy of Drosophila melanogaster and a set of terms to annotate data related to that anatomy. Extensive, well referenced textual definitions make it both a reliable and useful reference and ensure accurate use in annotation. Wide use of formal axioms allows a large proportion of classification to be automated and the use of consistency checking to eliminate errors. This increased formalisation has resulted in significant improvements to the completeness and accuracy of classification. The broad use of both formal and informal definitions make further development of the ontology sustainable and scalable. The patterns of formalisation used in the DAO are likely to be useful to developers of other anatomy ontologies.

Figure 1

Before and after refactoring. An example of incomplete classification, fixed by refactoring. (A) shows the classification of ‘mechano-chemosensory labral sensillum 8’ prior to refactoring. Many valid classifications are missing. Note also the erroneous classification of ‘external sensory organ’ as a type of ‘sensory organ cell’. (B) shows autoclassification of the same class after refactoring. Terms with equivalent class definitions are shown in green.

Figure 2

Auto-classification of sensory modality.(A) Classification under ‘detection of stimulus involved in sensory perception’. (B) Inferred classification of sensory neuron classes with sensory modality defined using the pattern: EquivalentTo neuron thatcapable_ofsome ‘detection of stimulus involved in sensory perception’ or one of its subclasses. (C) Classification and part relationships between subclasses of ‘sensory perception’ and subclasses of ‘detection of stimulus involved in sensory perception’. (D) Populating the auditory system. An auditory system neuron is defined as “any neuron that is capable_of_part_of (some) ‘sensory perception of sound”’. This property is directly asserted for ‘inferior ventrolateral protocerebrum IVLP-IVLP neuron’, and inferred for ‘auditory sensory neuron’, which is defined as “neuron thatcapable_of (some) ‘detection of mechanical stimulus involved in sensory perception of sound”’. Inference comes from the part relation between the two GO terms and a property chain stating that if X capable_of Y and Y part_of Z then Z capable_of_part_of Z. All auditory system neurons are asserted to be part_of (some) ‘auditory system’, an assertion that is inherited by all classes inferred to be subclasses of ‘auditory system neuron’.

Figure 3

Defining with images.(A) horizontal (B) frontal and (C) sagital sections through a standard reference Drosophila brain in which the lateral horn region is highlighted in purple with boundaries defined by the BrainName standard^c. We record a formal connection between an OWL individual representing the painted region and the class as: ‘lateral horn’ EquivalentTohas_reference_imagevalue ‘BrainName exemplar lateral horn’. (D) shows an image of single neuron that has been registered to the standard brain. Image analysis has determined overlap between the neuron and the region defined as lateral horn in the standard. Based on this, the neuron in the image has been annotated with the axiom: SubClassOfoverlapssome ‘lateral horn’. (Painting of standard brain by A.Jennet and K.Shinomiya. The single neuron image in panel C was derived from a neuron imaged by FlyCircuit [23] with registration and image analysis by MC and Gregory SXE Jefferis).

Figure 4

DAO content by system. Content of the DAO, divided by anatomical system.