Independent Continuant Subclasses: Really Disjoint?

[Stephen Larson]

Hello,

I have a question about the subclasses for independent contiunant.

The subclasses for Independent Continuants are:

FiatObjectPart, Object, ObjectAggregate, ObjectBoundary, Site,
SpatialRegion.

My question: these are defined as disjoint. Could someone explain how
they are disjoint? For example, I am working on a project to define an
ontology for neuroscience. It seems to me that depending on how you
look at it, an axon can be considered an Object, or an ObjectAggregate,
or a Site, or a Spatial Region.

Axon as Object:

If you were to lop off the axon from its cell body, you could weigh
it, put it in a box, and handle it as a self-connected entity.

Axon as ObjectAggregate:

Axons are composed of cell membrane, transmembrane proteins,
cytoskeletal proteins, myelinating cells that wrap around the outside.

Axon as Site:

Things can happen in an axon that don't happen in other parts of the
cell. Proteins can be transported along an axon, and certain
organelles are known to be present in different patterns in the axon
than in other parts of the neuron. In this sense, an axon is a
compartment of a neuron where things can happen.

Axon as a spatial region:

An axon has a three-dimensional shape, and inside of it, it has a
volume. When we observe axons in an electron micrograph, we are
basically looking at the axon as a region in space in the plane we
observe it in. We can measure its volume, and its presence defines a
spatial region that other cells cannot occupy.

While these subclasses make important distinctions, I am faced with the
practical problem of placing axon under just one of them. Each of
these aspects are important, particularly because I want to use this
ontology to do rule-based reasoning, and being able to consider these
objects as having multiple aspects provides important "semantic value".
It seems like I have 3 options:

1) Place axon under one of the categories, using a criteria of
"which aspect of an axon is the dominant one". This is good because it
provides for a simple hierarchy. However, by neglecting the other
aspects of what an axon is, it the definition loses semantic value and
explanatory power. Humans know that an axon can be thought from these
different perspectives...if we want to be able to make computers reason
about them, we probably can't get away with leaving this information
out.

2) Copy axon into Axon_Site, Axon_Spatial_Region, Axon_Object,
Axon_Object_Aggregate, and relate these instances by properties so they
hang together semantically. This view elevates the different aspects
of an Axon to the instance level. This allows for more semantic value,
but is impractical to maintain and is unintuitive.

3) Give axon multiple parent classes. This seems like the best
compromise, in that it is cleaner from a maintenance perspective, and
retains the semantic value. However, I understand that it is
discouraged to create ontologies that have subclasses with multiple
parents, because it leads to bad overall design of an ontology.
Moreover, I believe that the notion of these as "disjoint" means that a
subclass really cannot be more than one of them at the same time.

I imagine that this issue has come up before. Do others have
advice for how to deal with this? I would appreciate any help you can
provide.

Thanks,
Stephen

[William Bug]

Hi Steven,

Good to hear from you.

This is an excellent example of the difficult issues we must work through, if we are to effectively apply the OBO Foundry principles to complex, detailed descriptions of biological reality.  

I would say at the outset in the context our use of BFO in OBI, very similar questions have arisen, both in terms of the approriateness of "owl:distjointWith" and the exact distinction amongst the very independent continuants you discuss here - particularly ObjectAggregate and the relation between ObjectBoundary, Site, and SpatialRegion.  It's not clear to me whether we followed up on those discussions or not.  I'm partly expecting these issues will get a more complete fleshing out at the up coming PATO meeting, which is sort of the first nuts-n-bolts ontology meeting by those involved in the OBO Foundry effort since BFO.owl was released officially in September.  There was some discussion of these issues in the Sept. Common Anatomy Reference Ontology meeting, but for me at least - who wasn't present at that meeting - these discussions raised more questions than they answered and not all were direct reference to BFO, as it was rather new.

You are also touching on the between two of the major means of computational knowledge representation in biology - ontologies and models.  They have different purposes and different constraints.  Whereas models are particularly useful for producing simulations of the behavior of the modeled entities, ontologies are suited to supporting reasoning - whether DL-based or rule-based, though each of those two reasoning techniques have somewhat different requirements and constraints, as you were saying at the BIRN All-Hands meeting.  In the end, we certainly don't want ANY of these techniques - modeling (of which there are multiple techniques as well, of course) and ontology-based reasoning to produce contradictory observations, though any one may only be able to determine a subset of the aggregate knowledge derivable via all approaches considered as a whole.  Of course, there are others as well, such as machine learning/NLP/LSI on unstructured text, automated computer vision methods for discerning shape in 2, 3, & 4D digital series.  We hope to see these distinct - though overlapping - techniques will ultimately provide commensurate findings when applied effectively.

In regards to BFO in particular, I would also expect Holger, Andrew, Pierre, Barry, and Louis (see the "dc:creator" fields in the BFO.owl file - have each had to address some greater or lesser subset of these issues in various contexts to which BFO (SNAP + SPAN) and the Biodynamic Ontology have been applied over the past 2 - 3 years, as BFO has been evolving as a solid foundational ontology capable of supporting representation of real-world biomedical entities.

Here are a few comments that might simplify matters - just a little bit (at least, I hope they don't make things worse):

1) ObjectAggregate:

I believe in the case of an axon, you can rule this continuant out as an option.  Aggregate Objects from the point of view taken in ontological philosophy - if I understand this correctly - are, in fact objects that can have no existence outside their being a sum of non-distinct internal parts ("a mereological sum of separate objects possessing(sic) non-connected boundaries").  Good examples would be an army platoon which is an aggregate of all the soldiers of which it is composed and may be referred to in a meaningful way, without having to distinguish the individual soldiers.  Lipid bilayer would also fall into this category as a generic type of object, so long as the reasoning you were doing on an object defined in this way did not depend on the unique nature of the distinct lipid chains and glycerol phosphate functional groups present in the aggregate.  As a bilayer is considered an boundary forming object in biological systems, some might argue with this assignment of it as an aggregate object.  It depends on how it is being treated in the context of the observations being described, I suspect.  For instance FRAP imaging where you are studying the movement of molecules in the membrane bilayer certainly would require a definition of the lipid bilayer as more than just a physical boundary layer of non-distinct composition.  As an axon is composed of very distinct units, both in terms of its boundary layers and internal composition many of whose internal objects have very distinct, non-uniform distributions, I can't offhand think of any axon-related entity that would properly fit under this category.  Though I suppose one could make an convincing argument for both the microtubules and the collection of neurofilaments of playing a role best described as an "ObjectAggregate".  Though you certainly would not want to define an individual transmitter vesicle as an ObjectAggregate, you might want to describe the collection of transmitter vesicles clustered over a syanptic active site as such, though only if you could say with some degree of certainty, they in fact constituted a relatively uniform population in terms of their individual constituent components and transmitters - which in EM, you can certainly discern.  If I remember my structural biology of the synapse, there are generally sub-populations of active zones that do contain relatively uniform clusters of vesicles both based on diameter and histochemical composition.  There are other active zones clearly containing multi-modal distributions of vesicle types.  I believe such details are generally taken into account in typical, high-resolution synaptic models now-a-days.  But I digress...

2) ObjectBoundary:

Here again, I believe you would want to classify the axonal membrane as a boundary - and the various types of myelinating glial wrappings (Schwann Cell, oligodendrocytes) - as additional boundaries in specialized compartments.  These are compartments you would want to fully define via use of OWL ObjectProperties (see below).

This brings out an important distinction between structural definition of a compartment - the mereological definition of its boundaries and internal structural elements - and the functional definition of a compartment as being a space in which particular activities take place (I believe this would be a "Site" as defined in BFO).  In the case of the axon, as you point out, there are particular occurents found within the axonal space - the biological functions fast and slow antereograde transport and retrograde transport, for instance, which involve distinct components contained within the axonal space - kinesin and dynein respectively.  In fact, these complex processes will need to be defined as biological processes which are composed of specific molecular entities carrying out particular transport functions, where the the payload being transported are various other cellular components and molecules.

3) Site:

As you can see in the definition for this independent continuant, most of what comes under this universal type are geometrically defined locations ("characteristic spatial shape") often containing of complex collections of continuants disposed in non-contiguous orientations to one another.  This dependence on geometry is made clear by the fact an "owl:equivalentClass" has been defined for "Site" consisting of a  "UnionOf" Class Axiom of geometric elements of various dimensionality - e.g., 0D (point), 1D (line), 2D (surface/plane), 3D (volume).  Typically, lumenal cavities fall under this type, such as the interior of the gut or the ventricles of the brain, the interior of a lysosome, tight junctions, desmosomes, etc..   I've taken the liberty of adding in those final entities, because I think they also bespeak of the complexity which prompted your email.  I don't see any ontological reason why the interior of a lysosome should not be considered a volumetric "Site" if the gut cavity is considered as such.  Of course, we like to think of these cavities of consisting of a certain amount of void space, but in actuality, both the GI lumen and brain ventricles are most filled with fluid suspensions, and the amount of void cavity they contain relative to the "vacuolar"-like voids one can find in some sub-cellular compartments may simply be a matter of degree.  The tight junction is clearly a "Site" composed of the molecules (bfo:objects) such as claudin, occludin and a likely ObjectAggregate (microtubules - or the points of contact between claudin, occludin, and the microtubules).

In the case of an axon, I would say one should define "axonal_space" as a Site.  This independent_continuant is a component of the axon, and it contains the many elements found within the boundary structure - the axonal membrane.  The space defined by the sheath wrappings of Oligodendroglia would be another "Site" with a very complex, spiral volumetric shape - pinched-off, if you will, at each Node of Ranvier.  In fact, that complex spiral volume would really be just a part of the enclosing structure composed also of the neurolemma, incisures, glial nuclei, etc..  The overall entity is probably best described as a "bfo:object" with all of those constituent parts - including the sprial volumetric "Site" - though that's a site with little or no known function right now, I believe, though clearly the many molecular interactions between oligio membrane constituents and axonal membrane constituents are very important both in defining normal developmental processes such as Node and synapse formation, and in establishing how the axon responds to injury of various forms.

4) SpatialRegion:

Here too I think there is a very particular way in which this independent_continuant might be applied to an axon.  It was quite common in years gone by - possibly still today - for both physiologists studying the electrical excitability of axonal membrane and biochemists studying the biochemical machinery present in the axon, for researchers to use the Squid Giant Axon preparation to extrude the contents of the axonal space somewhat like one extrudes paste from a tootepaste tube.  In fact, both the fundemental physiological currents now ubiquitously described in most cells - not just neurons - and the early studies on the mechanics of axonal transport were observed using these techniques.  Based on the definition they provide in BFO from "SpatialRegion" and, in particular, the examples provided - "the space occupied by an appendix, the space that was occupied by an appendix prior to its removal, the space constitutive of a nasal cavity" - I would say the axonal_cavity remaining after one extrudes the axoplasm would rightly be classified as a BFO "SpatialRegion".  I would have to say the details of how "Site" & "SpatialRegion" are very subtle, when you get down to particular examples tied to specific instances.  Clearly, the examples given in BFO bespeak of more straight-forward entities that are more obviously distinct.  The example you pick, is a bit harder to map to these types.  It's probably the case that you would need to define the "Site" "axonal_space" for many of the instances you are trying to characterize, the "SpatialRegion" "axonal_cavity" would have an extremely limited domain over which it applied.  It would certainly be relevant when describing physiological experiments performed on an axon Squid Giant Axon whose extruded axoplasm was replaced with saline of defined composition.

Now - "axoplasm" is a sticky beast.  I'm not certain how you'd classify that.  Essentially, it's the collective components found within an axon, that once extruded actually hold together where the cytoskeletal components appear to form a matrix which gives the axoplasm the properties of a hydrated gel.

In general, I think "Object" is where you want to start for most of the continuant entities you describe in your email.  These - and some of the more abstracted entities I give above as examples - would then be properly inter-related via use of ObjectProperties.

It is clear - and this has been an area of considerable effort amongst the GO Consortia contributors such as the fly folks out of Berkeley (Chris Mungall, Suzi Lewis, et al.) and the Zebrafish folks out of Oregon and elsewhere - these complex relations need to be defined as transitive, inverse OWL ObjectProperties - e.g., "contains"/"is_contained_within", "transports"/"is_transported_by", etc.  As you know, each ObjectProperty brings with it a domain & range via which such complex biological processes can be gradually defined.  I believe there has already been considerable progress in this area which I hope to hear more about at the up coming PATO meeting just 10 days away.  Being able to describe instance-level phenotypic observations will require this level of complex relations.

On other note regarding the complexity of maintenance to which you refer.  You are certainly correct, if we are not careful regarding the means by which we represent these complex entities, we'll assemble an ontology requiring way too much manual effort to maintain, expand, and perform quality control on.  As you know, this is where OWL ObjectProperties play a critical role.  If properly employed, they will enable us to build out much of the ontological graph via inference, as opposed to manually describing all the assertions and constraints on all the entities, a process that will never scale.

In closing, I would also say I'm not completely certain "owl:disjointWith" really fits for all the interelations amongst these base "independent_continuent" types.  Using the examples in BFO, I think I can except one could create a formal definition of the "Site" "nasal cavity" that is in fact disjoint from the "SpatialRegion" "the space constitutive of the nasal cavity".  I'm much less certain it would be appropriate to present to a reasoner the assertion that the "Site" of "lysosomal digestion" is disjoint from the "SpatialRegion" defining the interior of a lysoszyme, but I probably don't fully understand how it is intended these continuants should be applied.

That's just my $0.02 on these issues.  Hopefully, others will be able to provide more authoritative answers to your specific questions regarding use of BFO entities.

Cheers,

Bill

[Stefan Schulz]

Hello,

I am new in in this list, let me introduce myself:
my name is Stefan Schulz, I am a Medical Informatics Researcher at Freiburg
University, Germany

Some comments to Stephen:

Hypothesis:
An axon is a FiatObjectPart according to the definition:
"An independent continuant that is part of an object
but is not demarcated by any physical discontinuities"
There is no bona fide boundary between the axon and the perikaryon
(body of the neuron), see:
http://mooreslore.corante.com/archives/images/axon.GIF
But would this not be a dependent continuant, since it is - by
definition - a part ?

> Axon as Object:
> If you were to lop off the axon from its cell body, you could weigh
> it, put it in a box, and handle it as a self-connected entity.

YES - but this is no argument againt FiatObjectPart

> Axon as ObjectAggregate:
>
> Axons are composed of cell membrane, transmembrane proteins,
> cytoskeletal proteins, myelinating cells that wrap around the outside.

NO , because these compounds are of very different types. The axons in
a nerve would constitute an ObjetAggregate
YES, because you may argue that it is composed of elementary
particles. As a consequence, every object would be an ObjectAggregate.
I have thought a lot (and published something) what pluralities really
are, and I concluded that whether something is an ObjectAggregate
depends on the glanularity of the ontology (both in terms of "grain
size" and classification criteria).

> Axon as Site:
>
> Things can happen in an axon that don't happen in other parts of the
> cell. Proteins can be transported along an axon, and certain
> organelles are known to be present in different patterns in the axon
> than in other parts of the neuron. In this sense, an axon is a
> compartment of a neuron where things can happen.

It may play the role of a site. I do not consider Site a sensible
upper level distinction. Everything can be a site of something. Site
is rather a reified relation than a upper level category

> Axon as a spatial region:
>
> An axon has a three-dimensional shape, and inside of it, it has a
> volume. When we observe axons in an electron micrograph, we are
> basically looking at the axon as a region in space in the plane we
> observe it in. We can measure its volume, and its presence defines a
> spatial region that other cells cannot occupy.

NO, because an object and its spatial region are two different entities.

Can anybody explain why FiatObjectParts and Sites are in this partition?

Kind regards
Stefan

[Cristian Cocos]

I can, at this moment, comment only on the distinction site/spatial region,
and I will try to put the matter in standard set-theoretic terms--that is,
in order to be understood by someone schooled in a standard 20th century
environment (which is not necessarily the view endorsed by the creators of
the BFO framework):

Think of axon-as-site along the lines of a set of spatial regions occupied
by the axon, or even as an equivalence class of such spatial regions. (Even
better, why not conceive the axon-as-site as a function t->set of (maximal)
spatial regions occupied by the axon during its whole existence (whatever
that may be).) I once witnessed Barry's explanation of why the mouth cavity
is a site and not a spatial region, and here's what (I recall) he said:
while spatial regions are static regions/subsets of spatial points, "your
mouth moves with you," as it were.

C

[Pierre Grenon]

Yes, the subclasses of Independent Continuant are disjoint.

A preliminary remark, SpatialRegion is not a subclass of Independent
Continuant in BFO. Do not confuse instances of Sites and instances of
SpatialRegion.

The reason why the subclasses of Independent are disjoint is due to
the fact that BFO is given in that layout under the assumption that a
specific level of granularity is fixed. Since the subclasses which are
given at the formal level essentially form a partonomy for Substance
(Boundary adds a tological flavor), whether an entity falls under one
of the subclasses or another will in general depend on the level of
granularity chosen for the ontolgy. (But it is not dependent on
granularity whether an entity falls under Independent_Continuant.)

The initial design behind BFO is that you would actually have as many
ontologies as you would want to represent granularity levels. This
much, I think, is something that the OBO foundry might illustrate --
for all I know about it.

As it happens, many domain specific ontologies are cross-granular and
it can be tricky to find a place where to put things.

One key to address this issue is to not try to make BFO do more than
it does. It gives a prototypical ontology at a formal level. Applying
BFO in a specific domain, however, does not necessarily mean fitting
or plugging domain specific classes under the BFO subclasses of
Independent_Continuant. It is very credible in the case of OBI for
instance taht there should not be a high level Aggregate or FiatPart
class. You could have your domain specific classes and do the
partonomy at their level. I.e. Axon, Aggregate_of_Axon, Axon_Part
etc...

This might be a way of having cross granular ontologies since it would
seem to put the granular load on classes rather than on taxonomies.

On 11/20/06, Stephen Larson <stephen...@gmail.com> wrote:
>
> Hello,
>
> I have a question about the subclasses for independent contiunant.
>
> The subclasses for Independent Continuants are:
>
> FiatObjectPart, Object, ObjectAggregate, ObjectBoundary, Site,
> SpatialRegion.
>
> My question: these are defined as disjoint. Could someone explain how
> they are disjoint? For example, I am working on a project to define an
> ontology for neuroscience. It seems to me that depending on how you
> look at it, an axon can be considered an Object, or an ObjectAggregate,
> or a Site, or a Spatial Region.

Yes, for BFO this denotes a choice of a granularity level (but for
SpatialRegion which denotes the fact that there is a mistake in the
implementation if that's where you got the notion that spatial regions
are under Independent_Continunant).

> Axon as Object:
>
> If you were to lop off the axon from its cell body, you could weigh
> it, put it in a box, and handle it as a self-connected entity.

Right. In that case, the cell is an Aggregate, right?

> Axon as ObjectAggregate:
>
> Axons are composed of cell membrane, transmembrane proteins,
> cytoskeletal proteins, myelinating cells that wrap around the outside.

Right.

> Axon as Site:
>
> Things can happen in an axon that don't happen in other parts of the
> cell. Proteins can be transported along an axon, and certain
> organelles are known to be present in different patterns in the axon
> than in other parts of the neuron. In this sense, an axon is a
> compartment of a neuron where things can happen.

You mean there's some room in it where some things which are not part
of the axon do stuff? If that's what you mean, yes, an axon is a part
of a site for those things (the hollow region -- or rather its filler
-- is another part of that site) .

I think, but that might actually be open, that it is not really
intended that an object by itself is a site, but virtually any object
with some surrounding (I mean something either internal or external,
surrounding might not be the best word) added to it will form an
entity which might be a site for some other.

> Axon as a spatial region:

no

> An axon has a three-dimensional shape, and inside of it, it has a
> volume. When we observe axons in an electron micrograph, we are
> basically looking at the axon as a region in space in the plane we
> observe it in. We can measure its volume, and its presence defines a
> spatial region that other cells cannot occupy.

That it has a volume (I presume you mean it is hollow?) makes it a
potential site (together with some of what's in there maybe).

That it has spatial characteristics makes it a spatial object, not a
spatial region.

It seems that avoiding multiple inheritance is one of the holiest
principles, I dare not speak for heresy. So, 3 is not available.

You want to look at what your problem and what you are suggesting not
as an issue of fitting a multiplicity of classes into a single
ontology but as an issue of working with multiple ontologies.

Each of the ontologies you have to manage is built according to your
solution-1, managing the multiple ontologies as a single ontology
amounts to your solution-2. The good news is that the properties
(binary relations) exist in BFO and that the classes which cause
troubles are actually defined via them.

In your boots, I would

1- simply not use the top level mereological classes (Aggregate, Parts, etc)
2- put Axon under the most likely (intuitively, I would say Object)
3- use relations such as hasPart, occupies, and their inverses in
order to link axons to their parts, to what they are parts of and to
what (sites) they occupy and to which they can be occupied as sites.

Notice now that by doing this you actually effect the same as 3. You
will not have the top level BFO subclasses of Independent_Continunant
as such, but you will be able to reconstruct as domains or ranges (or
restrictions of sorts) for the releveant relations something that
comes close to these classes with the disjointness requirement witheld
and allowing for multiple inheritance. These are set-theoretic
constructs, however, not classes that by the BFO scripture it would be
desirable to 'reify' (or 'name').

> I imagine that this issue has come up before. Do others have
> advice for how to deal with this? I would appreciate any help you can
> provide.

It definitely is a serious issue. If there are no provision in your
scheme to handle mutliple ontologies, you will have to alter, I think,
BFO using the spirit rather than the letter of the law as they say...
hope this helps.

cheers,
Pierre

> Thanks,
> Stephen
>
>
> >
>

[William Bug]

Hi All,

I'm having a hard time wrapping my mind around exactly what the recommendations are to Steven Larson.

Based on what I've been gleaning from the various replies:

1) Steven's intended application:

As he stated, he's working with EM cross-sections of neuronal tissue, where the goal is to take objects segmented (manually and semi-automatically) within the images, and identify them as components of axons and synapses (though his question focussed on the axon).  My sense is the "simplest" definition of axon that would work for this application is bfo:object.  The goal is to further define the details of how an "axon" is specified in the ontology (defined in OWL terms) in such as way so as to support rule-based reasoning beyond simply the assertion created when the segmented objects are specified as ontology class instances - e.g., the segmented objects are an instance of an axon boundary, or components of the myelination wrapping, or internal components within the axon.  It's not clear how best to apply these additional independent continuant classes to this process, beyond creating assertions via use of ObjectProperties to define mereological and functional relations on the axon as bfo:object.  Isn't this in fact what is being recommended for the more complex ontologies being assembled using OBO Foundry ontologies which at their base are expected to be using BFO?

2) Axon is an object with a Fiat Boundary (Stephan Schulz):

I would certainly agree with Stephan's observation that it is a matter of "fiat" as to exactly where the axon hillock is - and what it is composed of both structurally and functionally.  Functionally, it is considered from the neurophysiologist point to be the final integration point for all the combined excitatory & inhibitory membrane potential changes effected on the soma + dendrites and thus the titular site of origin of the action potential which travels down the axon.  There is experimental evidence that demonstrates this is in fact true - the AP appears to originate from the hillock, as if the density and characteristics of the Na+ & K+ in that spatial location are such that the hillock has the lowest threshold for originating the regenerative AP.  In that sense, the soma - though capable of generating an AP - acts an averaging capacitor for all the many +/- post-synaptic potentials.  Of course, this is a generalization.  In some specific neurons, the AP origin will differ a bit from this rule, based on differing distributions of the various V-dependent ion channels contributing to the AP threshold event.

Be that as it may, if you travel down from the hillock, it becomes even more a matter of fiat where you define the limits of the axon.  Many axons - though not all - will have branch points.  Others have recurrent collaterals that circle back and synapse on the axon to effect AP transfer function.  Finally, at the synapse end of the axon, given the dynamic nature of the intracellular scaffolding that defines the axoplasmic millieu, not only is the boundary between the axon and the synaptic terminal a bit diffuse, it also can change over time.

I would stress these are generalization, and any specific instance may vary from these general principles.

With all these facts in mind, an axon should certainly be considered an object with a fiat boundary.  Does that make it a FiatObjectPart according to the following definition:

"An independent continuant that is part of an object but is not demarcated by any physical discontinuities."

If so, how does one avoid creating an "axon" class with multiple parents?

3) ObjectAggregate:

Pierre seems to indicate both an axon and a cell could be grouped in the set of ObjectAggregates.  My understanding of ObjectAggregate is an object that literally only has no meaningful definition outside of being an aggregate of parts, where there is no relevant distinction made amongst the component parts.  The examples given for this class in BFO seem to support that definition:

"a heap of stones, a group of commuters on the subway, a collection of random bacteria, a flock of geese, the patients in a hospital"

In each of these examples, it is indicated that in the treatment of that ObjectAggregate, the individual parts are in fact indistinguishable.  My understanding both in terms of the set theoretic under-pinnings and the classic Leibnizian definition of an aggregate that this "indistinguishable" characteristic of the parts is important.  I don't believe it is required an aggregate defined in this way specify the cardinality of the constituent parts.

I could envision AggregateObjects being relevant to the characterization of certain entities within a cell - and within synapse in particular (e.g., the example given of the "aggregate" of vesicles found in proximity to some synaptic active zones).  I could also see  how in referring to organ parenchyma and some tissue components, it would be appropriate to have ObjectAggregates that consist of sets of cells considered in the aggregate where the individual cells are indistinguishable.

I don't see how any meaningful definition of either a cell or an axon would consider component parts as indistinguishable.

Many thanks for any more guidance you can provide on these points.

Cheers,

Bill

[Pierre Grenon]

> 1) Steven's intended application:
> As he stated, he's working with EM cross-sections of neuronal tissue, where
> the goal is to take objects segmented (manually and semi-automatically)
> within the images, and identify them as components of axons and synapses
> (though his question focussed on the axon).

I must have missed that statement.

> My sense is the "simplest"
> definition of axon that would work for this application is bfo:object. The
> goal is to further define the details of how an "axon" is specified in the
> ontology (defined in OWL terms) in such as way so as to support rule-based
> reasoning beyond simply the assertion created when the segmented objects are
> specified as ontology class instances - e.g., the segmented objects are an
> instance of an axon boundary, or components of the myelination wrapping, or
> internal components within the axon. It's not clear how best to apply these
> additional independent continuant classes to this process, beyond creating
> assertions via use of ObjectProperties to define mereological and functional
> relations on the axon as bfo:object. Isn't this in fact what is being
> recommended for the more complex ontologies being assembled using OBO
> Foundry ontologies which at their base are expected to be using BFO?
>

I will have to read your first point again tomorrow, it's a bit too
sophisticated for this time of the day although it seems like it's
calling for a "yes" answer. I think it is useful to better understand
the original question if that is indeed an accurate account of the
context. However, for one thing, I did not understand that a given
level of granularity was thought for and for another, I took at face
value Stephen's desire to have all these aspects of axons within a
single ontology. I also answered the question from the BFO standpoint
without actually worrying about the specifics of that particular
domain concerning which I know next to nothing.

>
> 2) Axon is an object with a Fiat Boundary (Stephan Schulz):

>

> Be that as it may, if you travel down from the hillock, it becomes even more
> a matter of fiat where you define the limits of the axon. Many axons -
> though not all - will have branch points. Others have recurrent collaterals
> that circle back and synapse on the axon to effect AP transfer function.
> Finally, at the synapse end of the axon, given the dynamic nature of the
> intracellular scaffolding that defines the axoplasmic millieu, not only is
> the boundary between the axon and the synaptic terminal a bit diffuse, it
> also can change over time.

The same goes for mountains, right? I think mountains are parts of the
earth but in some contexts --- if not most --- you can take them as
objects.

Now, whatabout a couple of lumps of playdo, one blue, one red, you mix
them a bit, do we have the same sort of configuration -- things with
vague boundaries. It seems to me there's a sense in which the mixture
is the objects and a sens in which it is an aggregate.

> I would stress these are generalization, and any specific instance may vary
> from these general principles.
>
> With all these facts in mind, an axon should certainly be considered an
> object with a fiat boundary. Does that make it a FiatObjectPart according
> to the following definition:
> "An independent continuant that is part of an object but is not demarcated
> by any physical discontinuities."

I'll pass on this one or I won't go to bed.

> If so, how does one avoid creating an "axon" class with multiple parents?

Well, suppose it does, killing some of the pqrents is one solution.

> 3) ObjectAggregate:
>
> Pierre seems to indicate both an axon and a cell could be grouped in the set
> of ObjectAggregates.

Actually, I said that if Axon is a subclass of Object, then Cell is a
subclass of Aggregate.
(I understood the claim according to which the axon is part of a cell,
correct me if that's wrong.)

Then, I said that it made sense to look at axons as aggregates because
they have parts which might be taken as objects, for example,
proteins. (Again, I am deferring the problem of the scientific
accurracy...)

The assumption was that these two claims would be made in distinct
ontologies, more precisely ontologies of that domain at different
levels of granularities.

> My understanding of ObjectAggregate is an object that
> literally only has no meaningful definition outside of being an aggregate of
> parts,

that much might be true enough to not launch here in qualifications
which would drive anybody insane

> where there is no relevant distinction made amongst the component
> parts.

That seems odd to me. From what follows I seem to understand that
instances of ObjectAggregate are aggregates of the same sort of
objects. The examples you cite indeed are such aggregates and they are
also the most intuitive and illustrative, but a flock of geese and a
duck is too an aggregate of objects, so is a heap of stone and wood
and so on and so forth. My desktop and my shoes form an aggregate of
objects.

> The examples given for this class in BFO seem to support that
> definition:
> "a heap of stones, a group of commuters on the subway, a collection of
> random bacteria, a flock of geese, the patients in a hospital"
>
> In each of these examples, it is indicated that in the treatment of that
> ObjectAggregate, the individual parts are in fact indistinguishable.

It is better not to infer such dramatic axioms from a small number of
examples. This suggests that either the comment should be amended so
as to deny your claim or some off the wall example should be added in
order to prevent similar conclusions.

> My
> understanding both in terms of the set theoretic under-pinnings and the
> classic Leibnizian definition of an aggregate that this "indistinguishable"
> characteristic of the parts is important. I don't believe it is required an
> aggregate defined in this way specify the cardinality of the constituent
> parts.
>
> I could envision AggregateObjects being relevant to the characterization of
> certain entities within a cell - and within synapse in particular (e.g., the
> example given of the "aggregate" of vesicles found in proximity to some
> synaptic active zones). I could also see how in referring to organ
> parenchyma and some tissue components, it would be appropriate to have
> ObjectAggregates that consist of sets of cells considered in the aggregate
> where the individual cells are indistinguishable.
>
> I don't see how any meaningful definition of either a cell or an axon would
> consider component parts as indistinguishable.

Me neither, although one could argue that being a cell part or being
an axon part would be good enough of a characteristics in which the
components would be indistinguishable. This is just to show that the
indistinguishability thing is ill conceived. Recall the group of
commutters in the subway example, are they actually indistinguishable
in any other characteristics than that of being parts of a group of
commutters in the subway? What if there are both males and females? Is
this no longer an aggregate? They might be all human, but in the same
way that cell components and axon components are all biological stuff.

> Many thanks for any more guidance you can provide on these points.

As I said, I'll try to give another read to your email, also to the
one you sent earlier.

cheers,
pierre

[Pierre Grenon]

Hiya,

> Here are a few comments that might simplify matters - just a little bit (at
> least, I hope they don't make things worse):
>
> 1) ObjectAggregate:
> I believe in the case of an axon, you can rule this continuant out as an
> option. Aggregate Objects from the point of view taken in ontological
> philosophy - if I understand this correctly - are, in fact objects that can
> have no existence outside their being a sum of non-distinct internal parts
> ("a mereological sum of separate objects possessing(sic) non-connected
> boundaries"). Good examples would be an army platoon which is an aggregate
> of all the soldiers of which it is composed and may be referred to in a
> meaningful way, without having to distinguish the individual soldiers.

I am not sure I understand this story about distinguishability.

Sorry the rest of the paragraph looses me entirely.

>
> 2) ObjectBoundary:
> Here again, I believe you would want to classify the axonal membrane as a
> boundary - and the various types of myelinating glial wrappings (Schwann
> Cell, oligodendrocytes) - as additional boundaries in specialized
> compartments. These are compartments you would want to fully define via use
> of OWL ObjectProperties (see below).

Well, strictly speak I think this is incorrect. The boundary of what
by the way? It takes some sort of naive physics type of approach in
order to consider a three-dimensional piece of matter as a boundary.
Although it is not -- as far as I know -- developed in any OWL type of
format, BFO has room for so-called external parts, which membranes are
excellent examples of. This is based on Smith's meretopological work
and has been partially used in the first order axiomatization of bfo.

> This brings out an important distinction between structural definition of a
> compartment - the mereological definition of its boundaries and internal
> structural elements - and the functional definition of a compartment as
> being a space in which particular activities take place (I believe this
> would be a "Site" as defined in BFO).

I do not understand the distinction made between structural and
functional definitions here.

> In the case of the axon, as you point
> out, there are particular occurents found within the axonal space

I'd say: some occurrents take place in that 'space', where 'space'
remains ambiguous between SpatialRegion and Site. In the spirit of
BFO, we would use SpatialRegion.

> - the
> biological functions fast and slow antereograde transport and retrograde
> transport, for instance, which involve distinct components contained within
> the axonal space - kinesin and dynein respectively. In fact, these complex
> processes will need to be defined as biological processes which are composed
> of specific molecular entities carrying out particular transport functions,
> where the the payload being transported are various other cellular
> components and molecules.

The processes are not composed of moleculra entities but they involve
them, the molecular stuff participates in these processes. There is a
risk when speaking of composition of having some sort of mereological
flavor added to that relation, whicvh would be wrong.

I'm not sure I understand why you bring this story about. Is there a
problem with this?

> 3) Site:
> As you can see in the definition for this independent continuant, most of
> what comes under this universal type are geometrically defined locations
> ("characteristic spatial shape") often containing of complex collections of
> continuants disposed in non-contiguous orientations to one another. This
> dependence on geometry is made clear by the fact an "owl:equivalentClass"
> has been defined for "Site" consisting of a "UnionOf" Class Axiom of
> geometric elements of various dimensionality - e.g., 0D (point), 1D (line),
> 2D (surface/plane), 3D (volume).

This doesn't make sense. Is this from the obi file or from the bfo file?

There is the issue of subsuming SpatialRegions under
IndependentContinuant in the BFO file but that further mistake is not
made there --- at least as far as I can tell. What is equivalent to
the union you are mentioning seems to be rather SpatialRegion. I think
even that is a simplification and is strictly speaking wrong but
that's not a very important problem. Making this equivalent to Site
would be a real issue.

It is definitely not the case that Site is the union of the named
subclasses of SpatialRegion.

Perhaps, howvever, this does not affect what you say below.

> Typically, lumenal cavities fall under
> this type, such as the interior of the gut or the ventricles of the brain,
> the interior of a lysosome, tight junctions, desmosomes, etc.. I've taken
> the liberty of adding in those final entities, because I think they also
> bespeak of the complexity which prompted your email. I don't see any
> ontological reason why the interior of a lysosome should not be considered a
> volumetric "Site" if the gut cavity is considered as such.

Me neither, was that suggested?

> Of course, we
> like to think of these cavities of consisting of a certain amount of void
> space, but in actuality, both the GI lumen and brain ventricles are most
> filled with fluid suspensions, and the amount of void cavity they contain
> relative to the "vacuolar"-like voids one can find in some sub-cellular
> compartments may simply be a matter of degree. The tight junction is
> clearly a "Site" composed of the molecules (bfo:objects) such as claudin,
> occludin and a likely ObjectAggregate (microtubules - or the points of
> contact between claudin, occludin, and the microtubules).
>
> In the case of an axon, I would say one should define "axonal_space" as a
> Site. This independent_continuant is a component of the axon, and it
> contains the many elements found within the boundary structure - the axonal
> membrane. The space defined by the sheath wrappings of Oligodendroglia
> would be another "Site" with a very complex, spiral volumetric shape -
> pinched-off, if you will, at each Node of Ranvier. In fact, that complex
> spiral volume would really be just a part of the enclosing structure
> composed also of the neurolemma, incisures, glial nuclei, etc.. The overall
> entity is probably best described as a "bfo:object" with all of those
> constituent parts - including the sprial volumetric "Site" - though that's a
> site with little or no known function right now, I believe, though clearly
> the many molecular interactions between oligio membrane constituents and
> axonal membrane constituents are very important both in defining normal
> developmental processes such as Node and synapse formation, and in
> establishing how the axon responds to injury of various forms.

I can't follow that sort of talk. There is no problem in introducing
an axonal_site and say that it's instances overlap with some instances
of axon. That much seems to be what is the general case and I'm not
sure more is needed. Of course, you can then add further
specializations of axonal_sites, based on which parts of the axon is
involved.

> 4) SpatialRegion:
> Here too I think there is a very particular way in which this
> independent_continuant might be applied to an axon.

Me too, that's the it doesn't apply way. :)

I think that what you say about spatial regions here probably applies
to sites and if you draw any distinction between sites and spatial
regions here they are distinction between different sorts of sites.

I have to stop here.

Cheers,
Pierre

[William Bug]

Sorry for all the text, Pierre.

I do appreciate your taking the time to provide this feedback, as I believe these question Steven asked are coming up in other venues.

Please see below for brief clarification.

Cheers,

Bill

On Nov 22, 2006, at 6:18 AM, Pierre Grenon wrote:

Hiya,

Here are a few comments that might simplify matters - just a little bit (at

least, I hope they don't make things worse):

1) ObjectAggregate:

 I believe in the case of an axon, you can rule this continuant out as an

option.  Aggregate Objects from the point of view taken in ontological

philosophy - if I understand this correctly - are, in fact objects that can

have no existence outside their being a sum of non-distinct internal parts

("a mereological sum of separate objects possessing(sic) non-connected

boundaries").  Good examples would be an army platoon which is an aggregate

of all the soldiers of which it is composed and may be referred to in a

meaningful way, without having to distinguish the individual soldiers.

I am not sure I understand this story about distinguishability.

Sorry the rest of the paragraph looses me entirely.

The idea here is if you are referring to an entity in such a way so that being able to distinguish the individual in the aggregate is not relevant, then the entity is truly an aggregate entity with whose relevance only exists in the reference to it being a collection of indistinguishable elements. Not that the elements themselves in some other context would not be fully distinguishable - just in this particular context it isn't relevant.  I'm taking this meaning strait from a discussion in one of Leibniz's philosophical works, but I've seen it referenced before - I thought in a set theory treatment of aggregate sets.  Cardinality of the individual objects in the aggregate may or may not be relevant, though as stated in the BFO definition for ObjectAggregate, they do consist of separate, distinct boundaries.  For instance:

"the collective weight of the commuters on the train was 1500 lbs or 681 kgs."

"the cost of 1 dozen croissant is 8 euros."

"typically, an aggregation of 15 or more neurotransmitter vesicles is found in close proximity on the presynaptic side of an active zones." (a fact quite often seen in the micrographs Steven is dealing with)

2) ObjectBoundary:

 Here again, I believe you would want to classify the axonal membrane as a

boundary - and the various types of myelinating glial wrappings (Schwann

Cell, oligodendrocytes) - as additional boundaries in specialized

compartments.  These are compartments you would want to fully define via use

of OWL ObjectProperties (see below).

Well, strictly speak I think this is incorrect. The boundary of what

by the way? It takes some sort of naive physics type of approach in

order to consider a three-dimensional piece of matter as a boundary.

Although it is not -- as far as I know -- developed in any OWL type of

format, BFO has room for so-called external parts, which membranes are

excellent examples of. This is based on Smith's meretopological work

and has been partially used in the first order axiomatization of bfo.

You are correct here, of course.  The membrane itself is not a boundary per se, but a physical object one side of which forms the limiting boundary for the interior of the axonal space.  In terms of demarcating the inside vs. the outside of the axon along its approximately cylindrical length, clearly the intracellular side of the axonal membrane would constitute the interior boundary, though the membrane itself constitutes a structural component of the axon. The exterior space is a bit more complex, and how you define it in conjunction with the various cellular constituents making up the myelin sheath will depend somewhat on what you expect to assert about the biological system, harking back to your comment regarding the granularity of the ontological relations being asserted.

 This brings out an important distinction between structural definition of a

compartment - the mereological definition of its boundaries and internal

structural elements - and the functional definition of a compartment as

being a space in which particular activities take place (I believe this

would be a "Site" as defined in BFO).

I do not understand the distinction made between structural and

functional definitions here.

There is the definition of a Site one might use to describe the structural location, then there would be the use of that Site in an OWL ObjectProperty such as "locatedIn" which would locate a particular biological process within that site.  That's the only distinction I was trying to make.  In both cases, the Site is the same, but the formal means one uses to locate function within it is different from the formal means used to define the Site itself.

In the case of the axon, as you point

out, there are particular occurents found within the axonal space

I'd say: some occurrents take place in that 'space', where 'space'

remains ambiguous between SpatialRegion and Site. In the spirit of

BFO, we would use SpatialRegion.

This doesn't appear to jibe with the description some one gave earlier in this thread regarding Barry Smith's distinction between SpatialRegion & Site.  If I remember it correctly - given in the context of the oral cavity - the distinction is the SpatialRegion is truly a location in space defined by coordinates that do not vary, whereas a site - such as the oral cavity - moves with you, as you move and thus occupies various SpatialRegions.  Both may defined via geometrical descriptions, though site would derive its description from the geometries of the structural components which define it, whereas a SpatialRegion can derive its definition simply from abstract geometrical constructs.  This jibes somewhat with the fact geometric objects of various dimensionality have been defined as owl:equivalentClasses of "SpatialRegion" in bfo.  In other words, the Site is the space defined by a collection of structural objects - a "relative" space if you will - whose actual coordinates may vary as those structural objects move about in space.  If this is the distinction used to differentiate Site from SpatialRegion, then clearly one would want to associated the biological processes taking place within an axon with a Site and not a SpatialRegion.

- the

biological functions fast and slow antereograde transport and retrograde

transport, for instance, which involve distinct components contained within

the axonal space - kinesin and dynein respectively.  In fact, these complex

processes will need to be defined as biological processes which are composed

of specific molecular entities carrying out particular transport functions,

where the the payload being transported are various other cellular

components and molecules.

The processes are not composed of moleculra entities but they involve

them, the molecular stuff participates in these processes. There is a

risk when speaking of composition of having some sort of mereological

flavor added to that relation, whicvh would be wrong.

I'm not sure I understand why you bring this story about. Is there a

problem with this?

Yes - again - you are definitely correctly.  I chose the wrong word here, though the point is to make it clear (something I clearly failed at doing) defining such biological functions (occurents) will require use of owl:ObjectProperties to fully express the functional details, as opposed to defining the process using a subsumptive graph - which is an error that has been made in constructing bio-ontologies in the past - as in the earlier iterations of GO.  This is not really a distinction I needed to clarify for Steven, however, as I know he understands this.

3) Site:

 As you can see in the definition for this independent continuant, most of

what comes under this universal type are geometrically defined locations

("characteristic spatial shape") often containing of complex collections of

continuants disposed in non-contiguous orientations to one another.  This

dependence on geometry is made clear by the fact an "owl:equivalentClass"

has been defined for "Site" consisting of a  "UnionOf" Class Axiom of

geometric elements of various dimensionality - e.g., 0D (point), 1D (line),

2D (surface/plane), 3D (volume).

This doesn't make sense. Is this from the obi file or from the bfo file?

There is the issue of subsuming SpatialRegions under

IndependentContinuant in the BFO file but that further mistake is not

made there --- at least as far as I can tell. What is equivalent to

the union you are mentioning seems to be rather SpatialRegion. I think

even that is a simplification and is strictly speaking wrong but

that's not a very important problem. Making this equivalent to Site

would be a real issue.

It is definitely not the case that Site is the union of the named

subclasses of SpatialRegion.

My mistake again.  I apologize.  I was looking at the file in raw XML format and read beyond the boundary of the definition of "Site".  The definition is taken directly from the bfo:Site definition:

"An independent continuant consisting of a characteristic spatial shape inhering in some arrangement of other continuant entities and of that which is enclosed in whole or in part by this characteristic spatial shape. Sites are characteristically entities at or in which other continuant entities can be located."

whereas the Collection of geometric shapes is defined as an owl:equivalentClass of bfo:SpatialRegion, which, as you point out, is a bit of a simplification, partly intended, I assume, to make the point I state above regarding the distinction between bfo:Site and bfo:SpatialRegion

Sorry for the error.

Perhaps, howvever, this does not affect what you say below.

Typically, lumenal cavities fall under

this type, such as the interior of the gut or the ventricles of the brain,

the interior of a lysosome, tight junctions, desmosomes, etc..   I've taken

the liberty of adding in those final entities, because I think they also

bespeak of the complexity which prompted your email.  I don't see any

ontological reason why the interior of a lysosome should not be considered a

volumetric "Site" if the gut cavity is considered as such.

Me neither, was that suggested?

Not directly in bfo, but I think the differing level of granularity sometimes throws people off.  The gut and the brain ventricles are macroscopic cavities.  No one would argue with that and most people think of them as such.  Sometimes the microscopic domains are not considered in this way, so it can be helpful to make that correspondence, I believe, though there is nothing in BFO or in the SNAP/SPAN articles that would lead you to think about the microscopic "Sites" any differently, so this confusion should not set in for those who have read the papers.

Yes - I agree.  All of the description that follows is still the description of a Site.  I think I have a more clear sense of the distinction between bfo:Site and bfo:SpatialRegion, though, as I say above, I know I am not the only one trying to work with BFO who has been confused by the distinction.

Perhaps it would be helpful to slightly amend the definitions, so as to make more clear SpatialRegion exists in coordinate system and can be defined by abstract geometric entities, where as a Site is defined as a relative location in reference to other continuants.  This is stated in the current definition "...consisting of a characteristic spatial shape INHERING IN (my emphasis) some arrangement of other continuant entities...", but I think there are many who don't quite pick up on this very important distinction based on the current formal ontological vernacular used to describe it.

Perhaps there is a need for a "BFO for Dummies" as well as a "BFO in a Nutshell".  ;-)

Cheers,

Bill

[Pierre Grenon]

> Sorry for all the text, Pierre.

You must have taken evening typing class...

> I do appreciate your taking the time to provide this feedback, as I believe
> these question Steven asked are coming up in other venues.

Well, you known, it's addicting they say.

>
> Here are a few comments that might simplify matters - just a little bit (at
> least, I hope they don't make things worse):
>
> 1) ObjectAggregate:
> I believe in the case of an axon, you can rule this continuant out as an
> option. Aggregate Objects from the point of view taken in ontological
> philosophy - if I understand this correctly - are, in fact objects that can
> have no existence outside their being a sum of non-distinct internal parts
> ("a mereological sum of separate objects possessing(sic) non-connected
> boundaries"). Good examples would be an army platoon which is an aggregate
> of all the soldiers of which it is composed and may be referred to in a
> meaningful way, without having to distinguish the individual soldiers.
>
> > I am not sure I understand this story about distinguishability.
>

> The idea here is if you are referring to an entity in such a way so that
> being able to distinguish the individual in the aggregate is not relevant,
> then the entity is truly an aggregate entity with whose relevance only
> exists in the reference to it being a collection of indistinguishable
> elements. Not that the elements themselves in some other context would not
> be fully distinguishable - just in this particular context it isn't
> relevant. I'm taking this meaning strait from a discussion in one of
> Leibniz's philosophical works, but I've seen it referenced before - I
> thought in a set theory treatment of aggregate sets. Cardinality of the
> individual objects in the aggregate may or may not be relevant, though as
> stated in the BFO definition for ObjectAggregate, they do consist of
> separate, distinct boundaries. For instance:
>
> "the collective weight of the commuters on the train was 1500 lbs or 681
> kgs."
>
> "the cost of 1 dozen croissant is 8 euros."
>
> "typically, an aggregation of 15 or more neurotransmitter vesicles is found
> in close proximity on the presynaptic side of an active zones." (a fact
> quite often seen in the micrographs Steven is dealing with)

I think I see the sort of things you are getting at here. But to me
this is threading on a thin line and it is not as much ontology as
linguistic. It makes sense to take talk about aggregates as one about
plural subjects, but whether you have these forms of speech does not
really matter to the notion of aggregation. It is nice to be able to
say something about the aggregate without worrying about its
constituents but it is precisely as useful and significant to be able
to tell which constituents are those of the aggregate in question. In
that sense, BFO is very much extensional, the aggregate is really
nothing more than the sum of its parts. As you rightly point out,
aggregates are entities, but then whether they fall under an
ontological category is not something which depends on the way we may
refer to them.

...

I think we agree on boundaries. At least we agree that the matter is complex.

...

> I do not understand the distinction made between structural and
> functional definitions here.
>
> There is the definition of a Site one might use to describe the structural
> location, then there would be the use of that Site in an OWL ObjectProperty
> such as "locatedIn" which would locate a particular biological process
> within that site. That's the only distinction I was trying to make. In
> both cases, the Site is the same, but the formal means one uses to locate
> function within it is different from the formal means used to define the
> Site itself.

Ok, I think I better understand. The trick with BFO is that some of
the classes are so complex to define that there is hardly anything
available beside what you would call functional definition, I gather.
For instance, the only definition of Site I can think of is to be the
range of the "occupies" relation (substantial occupy sites). In that
case the relation is simply a primitive.

>
> > > In the case of the axon, as you point
> > > out, there are particular occurents found within the axonal space
> >
> > I'd say: some occurrents take place in that 'space', where 'space'
> > remains ambiguous between SpatialRegion and Site. In the spirit of
> > BFO, we would use SpatialRegion.
>
> This doesn't appear to jibe with the description some one gave earlier in
> this thread regarding Barry Smith's distinction between SpatialRegion &
> Site. If I remember it correctly - given in the context of the oral cavity
> - the distinction is the SpatialRegion is truly a location in space defined
> by coordinates that do not vary, whereas a site - such as the oral cavity -
> moves with you, as you move and thus occupies various SpatialRegions. Both
> may defined via geometrical descriptions, though site would derive its
> description from the geometries of the structural components which define
> it, whereas a SpatialRegion can derive its definition simply from abstract
> geometrical constructs. This jibes somewhat with the fact geometric objects
> of various dimensionality have been defined as owl:equivalentClasses of
> "SpatialRegion" in bfo. In other words, the Site is the space defined by a
> collection of structural objects - a "relative" space if you will - whose
> actual coordinates may vary as those structural objects move about in space.
> If this is the distinction used to differentiate Site from SpatialRegion,
> then clearly one would want to associated the biological processes taking
> place within an axon with a Site and not a SpatialRegion.

This is a tricky argument we are getting into. You want to distinguish
various issues.

The first issue is that of the difference between two ontological
categories. The second issue is that of how to use them to do knowldge
rep.

With respect to the first question, it is one thing to find a
metaphorical or intuitive account of a distinction in order to make it
clearer, it is another to hard wire these intuitions in the concepts
themselves.

Here's one way I would suggest how to understand spatial regions. Take
a cylindrical container and fill it with marbles, these are special
marble they are homogeneous and infinitely dense, it is marble through
and through. There is 'empty space' between the marbles, however. It's
not absolute void, there's a medium such as air and so on. This empty
space is not a spatial region. The container with the marbles and the
empty space is a site.

Fill the container (with the marbles) with some slimy thing to the
point that we agree that there is no empty space. The slimy stuff is
special, it freezes everything, from now on, nothing can move (in
other words we are looking at a snapshot).

Now, this is the hard part of the thought experiment. Remove from the
cylinder filled with marbles and slime all the marbles. They are
anihilated on the spot with no reminder by some very special tool that
you only find in Stephen's lab.

What remains is just the container and the slime. There are islands of
nothingness in the slime where the marbles used to be. These iseland
are spatial regions.

Spatial regions are of another sort than substantial entities. Sites
are not. Substantial entities are located in space, and so are sites.
Sites can be occupied by other substantial entities. The notion of
site is very relative in the sense that it ought to be a site for
something...

Now, because sites are substantial entities, the problem of spatial
location is handled in bfo at the level of substantial for the
relevant portion of the ontology. Similarly, for perdurants which are
in a sense located in space, the most direct link is from them to a
given spatial region. Since sites are located in space, then, it is
almost immediate to make the link between perdurants and a site in
which they occur (roughly, they occur in the site that is located at
the region at which they occur). But even the occurrence of perdurants
in a region is in fine derivative on the spatial location of their
participants. That the particiants occupy a site --- in the scenario
you are interested in --- translates in terms of relations between the
participants' location and the site's location and this allows you to
link, eventually, the perdurants to a site.

What I'm getting at here is that in a FOL axiomatization of BFO, the
relation between a perdurant and a site is mere syntactic sugar. Of
course, in OWL, it's tough luck to make it so, but that's what I mean
when summoning the spirit of BFO.

...

> I chose the wrong word here,
> though the point is to make it clear (something I clearly failed at doing)
> defining such biological functions (occurents) will require use of
> owl:ObjectProperties to fully express the functional details, as opposed to
> defining the process using a subsumptive graph - which is an error that has
> been made in constructing bio-ontologies in the past - as in the earlier
> iterations of GO. This is not really a distinction I needed to clarify for
> Steven, however, as I know he understands this.

I see, I think. I'm afraid that in something like OWL there is no hope
of giving definitions for anything that is a bit complex other than
using an 'object property' indeed. Site is but another example.

>
> 3) Site:

...

> It is definitely not the case that Site is the union of the named
> subclasses of SpatialRegion.
>
> My mistake again. I apologize. I was looking at the file in raw XML format
> and read beyond the boundary of the definition of "Site". The definition is
> taken directly from the bfo:Site definition:
> "An independent continuant consisting of a characteristic spatial shape
> inhering in some arrangement of other continuant entities and of that which
> is enclosed in whole or in part by this characteristic spatial shape. Sites
> are characteristically entities at or in which other continuant entities can
> be located."
>
> whereas the Collection of geometric shapes is defined as an
> owl:equivalentClass of bfo:SpatialRegion, which, as you point out, is a bit
> of a simplification, partly intended, I assume, to make the point I state
> above regarding the distinction between bfo:Site and bfo:SpatialRegion

No need to apologize, it's not like it's cristal clear... There might
be some terminological dissonances between the owl file and my remarks
on BFO. Located in the comment for instance is to be understood in a
non technical sense (it is not referring to a relation in BFO since
the name of the intended relation is something like 'occupies' or
'occupied by'). Let me dispell another potential issue. I think that
the named subclasses of SpatialRegion are not shapes in the sense that
the comment on Site refers to a 'shape inhering ...'. I'm afraid
'shape' is ambiguous in English between 'shape property' and 'shaped
object'. Sites are shaped objects (but this is somewaht vacuous unless
you have some typology in mind). At least according to this comment,
in them inheres a 'shape property'. The named subclasses of
SpatialRegion are in a way divided according to shape (although the
genuine intent is to divide them according to dimensionality). So, you
could say that the named subclasses of SapatialRegion are types of
shaped regions but they are not shapes in the sense of 'shape
property' nor 'shaped objects' in the sense that in them inheres a
shape property (that's one further reason to distinguish them from
sites and other substantials).

...

> I think that what you say about spatial regions here probably applies
> to sites and if you draw any distinction between sites and spatial
> regions here they are distinction between different sorts of sites.
>
> Yes - I agree. All of the description that follows is still the description
> of a Site. I think I have a more clear sense of the distinction between
> bfo:Site and bfo:SpatialRegion, though, as I say above, I know I am not the
> only one trying to work with BFO who has been confused by the distinction.

OK. In my perception, much of these issues, however, are not unrelated
to terminological disagreements and to differring intuitions on how to
use the word 'space'.

Spatial regions are in BFO absolute, substantival, Newtonian space of
classical physics.

If you removed SpatialRegions the closest you get to is a relative
space which site are in a way --- but not precisely and not
limitatively --- intended to account for.

> Perhaps it would be helpful to slightly amend the definitions, so as to make
> more clear SpatialRegion exists in coordinate system and can be defined by
> abstract geometric entities, where as a Site is defined as a relative
> location in reference to other continuants.

Yes and no. It is a bit misleading to reduce the ontological
distinction to one of representation. The distinction between
coordinate system and relative (qualitative) representation is
something which is on the side of the ontological distinction.

> This is stated in the current
> definition "...consisting of a characteristic spatial shape INHERING IN (my
> emphasis) some arrangement of other continuant entities...", but I think
> there are many who don't quite pick up on this very important distinction
> based on the current formal ontological vernacular used to describe it.
>
> Perhaps there is a need for a "BFO for Dummies" as well as a "BFO in a
> Nutshell". ;-)

Something like this has actually been in the making and has recently
reached a mature stage, I think some people here at ifomis will soon
have some good news for you then.

cheers,
pierre

> Cheers,
> Bill
>
>
>

[Stephen Larson]

I want to thank Pierre, Bill, Stefan and Cristian, for their thoughts on this matter.  After discussing this matter with members of my lab, we have decided to make Axon a FiatObjectPart.  We will then examine the need to further define other aspects of axons such as "axonal_site", and combine them with properties, as we go further.

As a note, the exercise of trying to understand the subcategories of Independent continuant has been useful in clarifying our understanding of the entities we are defining.  I think that the BFO would benefit from having this kind of discussion included in a document on the website.  The references that I came across do a good job of explaining what the different subcategories mean individually, but I think that others would benefit from a simple explanation for why these subcategories are both disjoint and exhaustive.

Thanks,
   Stephen

[William Bug]

I would strongly agree with Steven regarding these discussions.

I think the BFO website you put out in September does a fantastic job providing links not only to the fundamental work on BFO but also several important applications.  The BFO Manual and "Related Material" pages are particularly valuable.  

Still, applying BFO to a broad range of highly specific biomedical domains at the level of very fine grained entities and their inter-relations, across a variety of spatial scales is still a complex process, even for those who have endeavored to absorb the knowledge in those valuable articles.  

It would be immensely helpful not only to document these sorts of discussions, but also to provide a means to easily find the work of others who've been using BFO - or other upper level ontologies (DOLCE and Simple Bio Upper Ontology, in particular) - to perform a variety of tasks in the biomedical domain.

I can only speak for myself here, but I expect proper application of BFO in holding with the OBO Foundry principles will remain a learning process for quite some time, though the pedagogical boost provided by the many projects now in the offing will take us users a great leap further toward the goal of developing large-scale, broad field ontology-based knowledge representation.

Cheers,

Bill