Integration with neuroanatomical atlases

Chris Mungall edited this page Jul 6, 2015 · 13 revisions

This page describes ongoing work to integrate Uberon with neuroanatomical sources of data

Background

Distinctions between ontologies and atlases

A neuroanatomical atlas is a map of brain regions or landmarks in 3D coordinates, typically obtained from a series of sections along an anatomical plane. An atlas can be used to derive a partonomy, a symbolic representation of the mereological relationships between landmarks. For example, the partonomy may record that the CA1 field is a part of Ammon's horn, because the set of voxels of the former is a subset of the set of voxels of the latter.

Most atlas partonomies are Jointly Exhaustive and Pairwise Distinct (JEPD). i.e. all parts are accounted for, and siblings do not overlap. In contrast, ontologies are capable of representing multiple ways of carving the brain. For example, pyramidal layer of CA1 can be a (layer) part of both the CA1 field as well as a (subdivision) part of the whole hippocampus pyramidal layer.

An atlas is typically derived from an individual brain at a snapshot in time. In formal ontology terms, these can be conceived of as being in the ABox (individuals, not classes). However, some atlases may be abstractions over multiple individuals.

An ontology is an abstraction over multiple individuals, providing generalizations that hold for all those individuals. An anatomical ontology represents classes of landmark. These classes are arranged in a generalization (subclass, subsumption hierarchy). The level of generalization varies depending on the ontology. Some ontologies are generalizations only for a particular species at a particular development stage (e.g. adult human). Uberon is a maximally general ontology, with all animals at all stages being in scope.

Atlases represent a single individual in a single species. Uberon is a multi-species ontology. Thus the class 'forebrain' is applicable across vertebrates, the class 'hippocampus' across mammals, and the class 'nervous system' across most of the Eumetazoa.

All axioms are assumed to be true for the class they are applicable for. If taxon-specific exceptions are to be encoded, we use the Evolutionary-variability-GCIs pattern

An atlas typically provides a single location for any named landmark. Classes in an ontology can be instantiated multiple places in any given brain. For example, in Uberon there are different subclasses of rhombomere, neural nucleus and white matter tract.

Alignment with atlases in Uberon

Allen Adult Mouse Neuroanatomy

The following is taken from the Uberon paper, and is still current:

Augmentation using Allen Brain Atlas

We downloaded the OWL version of the Allen Brain Atlas (ABA) and aligned it using the methods described above. We took advantage of the fact that the ABA, like most atlases, provides a non-overlapping parcellation, and derived spatial disjointness axioms to add to Uberon. The ABA is a partonomy that is represented in OWL as a subclass hierarchy. For every axiom in ABA of the form A DisjointWith B, we derived an axiom (part_of some A') DisjointWith (part_of some B'), where A' and B' are the Uberon equivalents of A and B. We represented this in the ontology using the spatially_disjoint_from shortcut relation (see below). For example, ABA contains the axiom: ``` ABA:HPF DisjointWith ABA:Isocortex ``` We used this to derive an axiom: ``` (part_of some UBERON:0002421) DisjointWith (part_of some UBERON:0001950) ``` where UBERON:0002421 has the label 'hippocampal formation' and UBERON:0001959 has the label 'neocortex'. These axioms were used to detect problems in Uberon, some of which could be traced back to source ontologies -- see, for example, [This item](https://sourceforge.net/tracker/?func=detail&aid = 3291162&group_id=76834&atid=1205376)

Since the paper was written, we have changed the prefix from 'ABA' to 'MBA' and use the official numeric identifiers rather than acronyms. We have also expanded to cover other Allen atlases

Bridge and composite ontologies

The ontology uberon-bridge-to-mba.owl contains bridging axioms between Uberon and the Allen Mouse Brain Atlas (MBA). It follows the standard pattern of all inter-anatomy-ontology-bridge-ontologies. Because MBA is mouse-specific, the axioms are of the form

 MBA:Forebrain EquivalentTo UBERON:forebrain and part_of some NCBITaxon:10090

This means MBA can be imported alongside Uberon, along with the bridge axioms. This will result in a lattice-like structure, e.g.

         Ubr:brain    ...
       /          \   /
      /            \ /
Ubr:forebrain   MBA:Brain
      \            /
       \          /
        MBA:forebrain

This can be compacted into a composite ontology that merges duplicative structures, leaving specific structures as leaf nodes or leafy branches.

For example, include the Allen primate brain atlas:

  • UBERON:0002301 ! layer of neocortex
    • (i) UBERON:0005394 ! cortical layer V
      • (p) PBA:10022 ! layer V of caudal orbitofrontal cortex (primate brain)

The composite ontology for MBA can be made available on request. For information on composite ontologies, see Multi-species-composite-ontologies

Note that the composite brain ontologies may be merged into composite-metazoan in the future

Integration with other Allen ontologies

Work is ongoing on the other ontologies - see #609

These are the current ontologies integrated. Each should have a corresponding bridge axiom

  • mba - adult mouse brain
  • dmba - developing mouse brain
  • hba - adult human brain
  • dhba - developing human brain
  • pba - non-human primate brain

HBA and lateralization

Note that for hba, the Allen structures are lateralized. For lateral structure, some combination of left, right or neutral is provided in the source. We choose neutral for the neutral uberon structures where available. Where hba provides l+r but no neutral, we select left. Of course, where the uberon class explicitly represents a left or right member, we select the appropriate one.

  • left: 167
  • right: 1

Integration with Neurolex

Neurolex contains regions from multiple sources/atlases. Some of these are mapped to a generic brain representation (which came from NIF_GrossAnatomy, in NIFSTD, which is all currently represented in Uberon).

Unfortunately Neurolex uses a semantic mediawiki RDF representation which is technically challenging to integrate. We have started a project to map neurolex to a more usable representation:

You can’t perform that action at this time.
You signed in with another tab or window. Reload to refresh your session. You signed out in another tab or window. Reload to refresh your session.
Press h to open a hovercard with more details.