Re: [Obo-discuss] Anatomical modifiers

[Chris M. <cj...@fr...>]

Hi Erik

Apologies for the length of this email, but it is a thorny and important
topic. We've discussed cross-products at length within GO. It would seem
that anatomical modifiers would actually be simpler than full blown
cross-products between the GO development terms and anatomical terms for
instance. However, your email brings up an important distinction that
needs to be made between structures and regions which complicates things.
I'll return to this later in the email.

Am I correct in saying your criterion for inclusion would be that the term
represents a morphologically discrete entity, and not a region within such
an entity?  Thus "dorsal fin" would be included (it's a whole fin, not a
region on a fin), but "anterior spinal cord" or "upper body" would be
excluded.

This seems like a reasonably rigorous criteria.

Presumably this would mean that "dorsal ectoderm" would be excluded from
the AO, as this is separated from "ventral ectoderm" by morphogen gradient
rather than a discrete boundary? I realise there may be some grey areas
where the notion of boundaries is slightly mushy, but this seems like a
clear case of exclusion by your rules.

Removing "dorsal ectoderm" from an AO will take away the ability to make
statements such as "tracheal primordium develops_from dorsal ectoderm".
This is a statement of general biological truth (for some species anyway)
and not a piece of data, and hence belongs in an ontology, not in
instances/annotations. You need to have both the structure "ectoderm" and
the region "dorsal ectoderm" in the AO to do this (well, you don't
actually need to, but it's easier).

There may be similar issues for the AO partonomy with removing
"anterior spinal cord", it depends how detailed your partonomy is.

It would seem to be a good idea to encourage all OBO anatomy ontologies to
adopt the same criteria for term inclusion/exclusion.

Regardless of these criteria, it is definitely a good idea to develop
a seperate ontology of spatial qualifiers. This should be
species-neutral, although species-specific extensions may be required
(eg thoracic).

Using the forthcoming cross-product DAG-Edit plug-in, it will be possible
to easily generate qualified terms. In addition, the "class definition"
(ie the cross-product) will be saved which will make it easier to
automatically place cross-product terms in the DAG. It will also make it
easier to add filters to browsers such that "impure" qualified terms are
dynamically excluded (to do this now would require text parsing of the
terms). In effect the "Frankenstein monster" you fear should be easily
containable. I favour eventually placing these qualified terms in the
ontology, although you may prefer to only use the qualifiers in
annotations for the present time.

However, there is one important subtlety that needs to be addressed before
we develop the spatial qualifier ontology for use either with
cross-product terms or with annotations.

In some circumstances, the meaning of combining a spatial qualifier string
with a structure string yields a string that stands for an anatomical
structure:

  A: qualifier + structure = structure (eg "dorsal fin")

In other circumstances, the combination yields a region:

  B: qualifier + structure = region (eg "dorsal ectoderm")

One of the cardinal rules of ontology design is that ontological
categories must not be crossed by an 'is_a'. Thus, "dorsal ectoderm" would
NOT be placed as an is_a child of "ectoderm" - the relationship is
spatial. "dorsal ectoderm" belongs as an is_a child of "dorsal region"
(currently it is an is_a orphan in the fly anatomy). In these cases the
adjective "dorsal" is not serving to qualify the meaning of the noun
phrase - it is changing the meaning from a structure to a region.

Similarly, an "X primordium" would not be placed as an is_a child
of "X", but as an is_a child of "primordium" (or "Y primordium", where X
is_a Y).

Contrast this with "dorsal fin". In this case, "dorsal" qualifies the
meaning of "fin" ("dorsal fin is_a fin").

This has serious implications for how the proposed spatial qualifier
ontology is used. I would argue that for the majority of cases what we
really want is a core ontology of spatial regions, rather than an ontology
of qualifiers.

We actually have 3 categories or orthogonal ontologies here:

1. qualifiers (eg dorsal, ventral)

   these are terms that have no spatial extent or physical existence
   in themselves. They can be applied to (2) or (3) below.

2. regions (aka dependent parts)

   a region has 3d spatial extent, but depends for its existence on
   a structure; the spatial extent is relative to a structure.

   the region ontology has perhaps only one "pure" term: region

   it can be combined with 1 to make qualified regions; eg:
   "dorsal region", "ventrolateral region".

   these combinations can then be applied to (3) to make fully specified
   regions, located relative to a structure; eg:
   "upper region of body", "dorsal region of ectoderm" (aka "dorsal
   ectoderm"). These terms belong here in (2) and not in (3)

3. structures (eg body, fin, ectoderm)

   a distinct morphological structure with spatial 3d extent that
   does not necessarily depend for its existence on other objects

   pure terms would be: body, fin, ectoderm, spinal cord

   structures can be qualified by (1) to differentiate them from
   other structures of the same name; eg
   "dorsal fin" "right leg" "left ventricle"

   however, there is no such _structure_ as a "dorsal ectoderm" -
   it is the region on a structure

   structures can be the location for a region (eg ectoderm is the
   location for the "dorsal ectoderm" region).

(There are also "surfaces" which can be treated analagously to
"regions", although they would of course have different qualifiers,
such as inner/outer; the same applies to aggregates such as "systems")

If we ignore this subtlety and implement a hacked solution, it will mean
that it will be impossible for a computer to tell the relation between
"dorsal fin" and "fin" is "subclass" and between "dorsal ectoderm" and
"ectoderm" is "part_of"/"located_in".

Here is how I propose we do this:

SolutionA:

We create both ontologies 1 and 2 above. In fact, the region ontology
can be generated automatically from 1.

When creating either cross-product annotations or cross-product terms
for use with simple annotations (we can map between the two), the
annotator or curator selects the correct ontology 1 or 2 depending on
whether the desired term is a structure or a region.

Would this seem excessively onerous? Would curators/annotators be happy
with this, or is this too abstract? How likely is it that errors would
arise by either accidentally making "dorsal ectoderm" a structure or
"dorsal fin" a region? Is this likelihood much higher if the
cross-products are made at annotation time as opposed to curation time?
(this is one reason for keeping things centralised in the ontology -
annotators are more likely to find different ways of constructing the same
concept and similarly the same cross product specifying different
concepts: compare "cell homeostasis" with "B-cell homeostasis")

Another solution (Solution B) is to omit 1 and only have regions and
structures. "dorsal fin" would be a cross-product involving the
primary term "fin" and the differentium "dorsal region" (of whole
organism). Constrast this with "dorsal ectoderm" which would have
primary term "dorsal region" and differentium "located-in
ectoderm". The primary term determines the category.

Again, errors will occur if this is used incorrectly. If "ectoderm" is
used as the primary term in a cross-product with "dorsal" it will
yield a concept that has no counterpart in the real world. If this is
hard for people to get their heads around then errors will occur.

For those interested in the formal ontological principles behind this,
Barry Smith has written extensively on this subject. The 3 categories
above correspond to some of the upper layers in his Basic Formal
Ontology model (what he calls "dependent entities", "fiat parts of
substances" and "substances"). I'm afraid I don't have the actual
reference at hand. It is also worthwhile looking at the upper level
subclass hierarchy of the FMA, which makes similar divisions. I think
we would do well to do the same.

Regards
Chris

On Mon, 21 Feb 2005, Erik Segerdell wrote:

> Greetings,
>
> ZFIN curators have begun large-scale curation of gene expression data,
> and we plan to curate mutant data following the phenotype ontology
> (PATO) syntax. We use the Zebrafish Anatomical Ontology (AO), which we
> are actively developing, for our annotations.
>
> We find that we frequently need to "modify" an anatomical structure in
> certain ways to more accurately describe the data. For example,
> expression may occur in the anterior part of structure, but not the
> posterior region. We would like to capture these distinctions.
>
> The most obvious solution is to simply add new terms with regional
> distinctions to the AO , e.g. "anterior spinal cord" and "posterior
> spinal cord" as child terms of "spinal cord". Unfortunately, this may
> lead quickly to a Frankenstein monster of an ontology being unleashed
> on our unsuspecting users! Also, we wish to maintain a certain
> anatomical purity in the ontology, and include only items that are, in
> and of themselves, observably distinct structures. This leads to many
> grey areas, of course, but arbitrarily creating a bunch of anterior and
> posterior structures would violate the spirit of what we want the AO to
> be.
>
> In some cases, the name of a structure already conveys directional
> information, e.g. "dorsal fin", to distinguish it from a similar but
> entirely separate structure ("tail fin"), and our AO will contain
> separate terms for these. Rather, our concern is to distinguish, within
> a structure, regions that cannot appropriately be considered
> observable, morphologically discrete entities.
>
> To this end, we want to use a controlled list of modifiers that we can
> optionally tag on in front of a structure in an annotation.
>
> A group of us at ZFIN have discussed this need and we are not sure
> whether this is a curation detail that requires little more than an
> extra data column in a table in our database - or whether this is
> something to address at the level of the ontology itself, in some way?
> Also, are there hidden complications to this approach, as in the long
> run we plan to have links and translation tables between our ontologies
> and annotations and other model organism databases?
>
> Finally, if anyone has already developed a controlled vocabulary of
> this sort, we would certainly prefer to share terms. (One can imagine a
> modifier ontology, since these terms should be defined and can have
> synonyms.)
>
> (BTW, I did bring up this issue in a post to the obo-phenotype
> list-serve on 8 Dec 2004 and received feedback from Chris Mungall...
> but in retrospect it was bit PATO-centric of me to mention it in that
> forum, as it's turning out to be a more general problem for ZFIN.)
>
> Our draft list of modifiers includes:
>
> apical
> basal
>
> anterior
> posterior
>
> dorsal
> ventral
>
> distal
> proximal
>
> medial
> lateral
>
> deep
> superficial
>
> We are considering two more modifiers to deal with a developmental
> timing problem:
>
> primordium_of
> remnant_of
>
> e.g., we may want to record "primordium_of otic placode" for expression
> occurring in the part of the ectoderm fated to be the otic placode, or
> we may want to say "remnant_of yolk syncytial layer" in a case where an
> author claims expression in a disappearing structure beyond the
> developmental stages at which that structure is commonly defined to
> exist. In such cases, we would like to capture the info without adding
> an inappropriate new term to the AO or unnecessarily expanding the AO's
> definition of when these embryonic structures appear and disappear.
>
> We appreciate any comments and insights. Thanks!
>
> Erik Segerdell
> Scientific Curator/Nomenclature Coordinator
> Zebrafish Information Network (ZFIN)
> 5291 University of Oregon
> Eugene, OR 97403-5291
> Tel 541.346.3989
>
> Email se...@uo...
> URL http://zfin.org
>
>
>
> -------------------------------------------------------
> SF email is sponsored by - The IT Product Guide
> Read honest & candid reviews on hundreds of IT Products from real users.
> Discover which products truly live up to the hype. Start reading now.
> http://ads.osdn.com/?ad_id=6595&alloc_id=14396&op=click
> _______________________________________________
> Obo-discuss mailing list
> Obo...@li...
> https://lists.sourceforge.net/lists/listinfo/obo-discuss
>