Consider whether range of 'has input' should be broader

[DanFaria]

While it is clear that all processes must involve a material entity in some capacity, the input of a process (as one normally understands the meaning of input) need not be a material entity. For example, in the digital processes of cropping an image file or sorting a tabular file there is a material entity involved (the computer where these processes take place) but one would not normally define the computer as an input of the process (formally, it is, but it is of little practical use to ontologically define digital processes as changes of state in a computer). Rather, the intuitive (and useful) usage of input in these examples would be the image file and the tabular file. Would it not make sense to relax the domain of 'has input' to 'continuant' in order to cover digital processes?

[cmungall]

My answers may also apply to #316

I personally don't spend much time modeling informational processes so I have more questions than answers at this stage. Perhaps this could be farmed out to the IAO crowd

What are the participants of a digital process? Is it the hardware, electrons flowing, etc? Or is it the information entities? Or is it both? Perhaps there are two processes, a 'material process' and an 'information process'? I feel we need to get a coherent picture of this first of all before making changes that may need to be reversed later.

@jamesaoverton is this something OBI people think about a lot?

[DanFaria]

From the perspective of someone trying to model experimental processes, both physical and digital, my view is that there are indeed two ways to look at the process---'material process' vs 'information process'---though in practice the latter is more useful than the former for most ontological applications.

The only scenario in which I can contemplate caring about the state of digital hardware at the physical level is an ontology devoted specifically to modeling computer operations (say, for a CPU manufacturing company).

In general, what we care about knowing for digital processes is the 'information process', i.e., what were the input and output files, what programs were executed or what instructions thereof. In particular, if we wanted to report an experimental procedure that includes digital analysis processes, then this is what we would need to report in order to enable others to understand and even reproduce the experiment.

Bear in mind also that, for the most part, we assume that the physical machine in which digital processes take place is of little or no concern. In computer science, we usually report that the process 'occurs in' a machine with given hardware specifications and operating system, but that is the extent to which we care about it---we assume that we could replace the machine used with any other machine with comparable characteristics. And in other scientific domains, such as the life sciences, the hardware specifications are often omitted altogether. We expect that, regardless of hardware, the same program with the same parameters and input data would yield the same output.

In any case, my suggestion with respect to relaxing the domain of 'has input' would enable RO to be used (in conjunction with BFO) to model digital processes in the way we most often care about---as 'information processes'---without compromising its ability to model them as 'material processes' still. Thus, the only real issue is whether you are willing to extend the scope of RO to contemplate 'information processes'.

[jamesaoverton]

I think @DanFaria will be interested in these two object properties from OBI:

• 'has_specified_input' http://purl.obolibrary.org/obo/OBI_0000293
• 'has_specified_output' http://purl.obolibrary.org/obo/OBI_0000299

They are both subproperties of 'has participant' and inherit the domain 'occurrent' and range 'continuant'. The "specified" part indicates that that some plan is involved, so in practise the domain is almost always 'planned process'.

We have three important subclasses of 'planned process' that differ by the ranges of the inputs and outputs

• 'material processing' has specified input some 'material entity' and has specified output only 'material entity' (making the output a 'processed material')
• 'data transformation' has specified input some 'data item' and has specified output some 'data item'
• 'assay' has specified input some 'material entity' and has specified output some 'data item', thus moving from the realm of material thing to the realm of information

OBI sticks to the "information level" with 'data transformation', as @DanFaria is describing. We have some terms for talking about the physical substrate of information that @cmungall raised, such as IAO 'material information bearer' and the 'concretizes' relation, but that's rarely the level that we care about.

To the best of my knowledge, these OBI terms predate RO 'has input'. OBI uses several RO Core relations, but not RO 'has input'.

If @DanFaria is only really interested in 'planned process'es, then maybe this OBI terms are suitable.

[cmungall]

I agree with both your analyses, thanks. The solution for Dan is to use this relation.

The split between having "material/phsyical/biological" stuff in RO and "information-y" relations in IAO and OBI is not ideal but is not an immediate concern. The terminological mismatch in having 'input' sound like the parent of 'specified input' is also not ideal but is a terminological issue rather than logical one.

More concerning is the disconnect between the BFO2 reference and the OBI/IAO/RO modeling. James, although you do not state a range for either of the properties you link to, it is implicitly inclusive of GDCs (and this is not contradicted by the "continuant" range of the parent in RO).

However, the BFO2 reference document states the range of has-participant to be "independent continuant that is not a spatial region, specifically dependent continuant, generically dependent continuant".

This is clearly a problem. It looks like BFO2 has a very materialistic view of processes. Given that this inconsistency was introduced after the long-standing OBI relations, I am sure there must have been some discussion of this (BFO wouldn't just introduce something that breaks things without consulting their users, would they?) - am I missing something here?

Maybe something to discuss at obo-core workshop at ICBO, cc @bpeters42

[DanFaria]

Thanks for your input @jamesaoverton.

Alas, using OBI is not the solution I am looking for. I am looking for an upper level ontology to use as the basis for an ontology about experimental pipelines (focused on the plant domain). I was using BFO as such before BFO relations were moved to RO, and started looking into using RO as well after they were.

If the upper-level structure of OBI were an independent ontology or part of BFO & RO (wouldn't it make sense to do so?), I would certainly be using it. But as is, it is both too specific and too extensive for this purpose, and because it overlaps with the ontology I'm working on, importing it would lead to terminological and possible logical conflicts. I do plan on cross-referencing to OBI whenever applicable, though.

My issue wrt BFO & RO is clearly tied to the fact that they have a very materialistic view of processes, as @cmungall put it. That it clashes with OBI is all the more reason to look into whether a broader view wouldn't be better for an upper-level ontology.

[cmungall]

Alas, using OBI is not the solution I am looking for. I am looking for an upper level ontology to use as the basis for an ontology about experimental pipelines

I think this can be fairly easily resolved. The OBI relations could be included as part of the RO release, or OBI and RO could work together to obsolete the existing IDs and mint RO IDs (this would obviously cause churn).

But we need to resolve the ontological issue first, whereby there is currently a cryptic incoherency in OBI

[jamesaoverton]

@cmungall The ontological problem is a serious one, and it's not new: see #3. I had not thought of it in terms of these OBI relations, but saying "there's a cryptic incoherency in OBI" is completely misleading. RO 'has participant' and OBI 'has specified output' predate the BFO2 spec by years. OBI is using RO 'has participant' correctly.

I think that the term reuse "problem" is a serious misunderstanding of how OBO works. I'm hearing @DanFaria say "OBI has the right terms, but I only want to use BFO and RO", and then @cmungall reply "OK, we'll obsolete the OBI terms and make new ones in RO". I cannot disagree more strongly. In OBO we reuse each other's terms, and we make it as easy as possible to mix-and-match.

@DanFaria I don't understand the reasons you've given for not using OBI. Of course OBI is not an "independent ontology" -- it extends BFO, uses RO, and uses many other OBO ontologies in order to form a rich network of axioms. That's the central goal of OBO.

If the main OBI release doesn't suit your needs, consider using OBI Core which includes about 100 of our upper-level terms, and was designed for just such a purpose. You can also pick the subset of terms that you need from OBI, BFO, RO, and others. That's how OBO works.

In any case, I encourage you to get in touch with OBI developers on our weekly calls, mailing list, or issue tracker: http://obi-ontology.org. We are very interested in experimental pipelines.

[DanFaria]

@jamesaoverton I understand things are viewed differently in OBO, but I'm developing an OWL ontology, and to comply with OWL best practices, I would have to import the whole OBI to reuse its relations. This would mean importing a number of classes that potentially clash with my ontology's classes. That is why I was looking to import an upper-level ontology only.

I was not previously aware of OBI Core and will look into it. It may serve my purposes.

[jamesaoverton]

@DanFaria There are many tools that you can use to extract a subset of terms from an OWL ontology (or set of ontologies) while preserving logical axioms. This is daily practise in the OBO community, but there's nothing OBO-specific about reusing parts of OWL ontologies. I recommend our ROBOT tool, which provides an extract command: http://robot.obolibrary.org/extract.

[phismith]

Can you explain in what sense BFO has a materialistic view of processes Thinking processes Changes in stock exchange processes Changes in temperature processes Etc etc have all been modeled using BFO and no one this far has registered any problems BS

On Thu, Apr 11, 2019 at 12:56 Daniel Faria [email protected] wrote:

Thanks for your input @jamesaoverton https://github.com/jamesaoverton.

Alas, using OBI is not the solution I am looking for. I am looking for an upper level ontology to use as the basis for an ontology about experimental pipelines (focused on the plant domain). I was using BFO as such before BFO relations were moved to RO, and started looking into using RO as well after they were.

If the upper-level structure of OBI were an independent ontology or part of BFO & RO (wouldn't it make sense to do so?), I would certainly be using it. But as is, it is both too specific and too extensive for this purpose, and because it overlaps with the ontology I'm working on, importing it would lead to terminological and possible logical conflicts. I do plan on cross-referencing to OBI whenever applicable, though.

My issue wrt BFO & RO is clearly tied to the fact that they have a very materialistic view of processes, as @cmungall https://github.com/cmungall put it. That it clashes with OBI is all the more reason to look into whether a broader view wouldn't be better for an upper-level ontology.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/oborel/obo-relations/issues/315#issuecomment-482067571, or mute the thread https://github.com/notifications/unsubscribe-auth/AH6qN6__mgv36BT-0-Te7RxEZx2nfOYWks5vfxTWgaJpZM4ccEq1 .

[DanFaria]

@jamesaoverton You misunderstand me. I am well aware that it is possible to extract a subset of terms from an OWL ontology and reuse them in another ontology. However, I believe that doing so is not a good OWL ontology development practice. We end up with the same entity being defined in two different ontologies which defeats the purpose of using URIs, and can lead to problems later on if one ontology updates this entity and the other doesn't. I prefer the alternative of defining the entities I need under the namespace of my ontology and cross-referencing to other ontologies when adequate.

@phismith The range of 'has input' being restricted to 'material entity', as per my first post above, precludes the definitions of digital objects as inputs of information processes, as further detailed in my second post. If this were the only issue, it would still be possible to model information processes in BFO, using 'has participant' or declaring sub-properties thereof (as in OBI), but according to @cmungall above:

However, the BFO2 reference document states the range of has-participant to be "independent continuant that is not a spatial region, specifically dependent continuant, generically dependent continuant".

[jamesaoverton]

@DanFaria Ok, I understand the point you're making about the downsides of reuse. In the OBO community we're developing a distinction between "full" and "base" OWL artifacts that's designed to mitigate those downsides. You can follow our discussion here, if you're interested: https://github.com/INCATools/ontology-development-kit/issues/194

I would also like to hear @phismith's thoughts on 'has participant' and its range in RO and BFO2.

[bpeters42]

Completely agreeing with James: There seems to be a major BFO2 disconnect to RO and subsequently OBI/IAO. We have always modeled 'data transformation processes' and the like as acting on information content entities. For BFO2, we have implemented the classes, but are not using the relations, so we never noticed the inconsistency you point to. If BFO2 is settled on 'has participant' to be limited to materials, we could create another property such as 'has generalized participant' (never mind the sucky name for now), which would cover both material entities and their dependents (both specific and generic). It does seem a good topic for the OBO core workshop and/or follow up to the RO one.

[cmungall]

I put this down for our May 7 call, if we have time. Hopefully you can make it @phismith.

@DanFaria if you are on the RO list you should receive an invitation

[alanruttenberg]

BFO FOL has domain of participates in (independent continuant and not spatial region) or specifically dependent continuant or generically-dependent-continuant. Axiom imply that bearers of the dependents also participate

[cmungall]

Thanks. On the call you also mention some axioms that propagate e.g. if X participates then Y inheres in Y then Y participates?

[pbuttigieg]

Axiom imply that bearers of the dependents also participate

This is very useful - I can see this being an asset for handling molecular ecology phenomena which are based on information transfer (e.g. some sub-phenomena of genetic flows)

[cmungall]

BFO FOL has domain of participates in (independent continuant and not spatial region) or specifically dependent continuant or generically-dependent-continuant.

Could this not be simplified to "continuant and not spatial region"? The current public BFO2 is:

• continuant
  • IC
  • GDC
  • SDC

If we assume this is not closed then we can simplify. Of course, it is not closed and BFO permits other subclasses of continuant, but why restrict these from being participants?

Axiom imply that bearers of the dependents also participate

This is a very powerful axiom, potentially very useful. But it may also potentially take people by surprise. The entailments may be unintentional for some. I recommend some procedure in which existing BFO users can be forewarned and discuss axioms like this.

[cmungall]

@alanruttenberg can you specify what the new BFO axioms are? We'd like RO to be as consistent as possible. E.g. is the propagation only in one direction (bearers -> dependents, or also dependents -> bearers)?

What is the best way to be forewarned of changes coming to BFO? Should these be announced on the BFO list?

[nlharris]

@alanruttenberg Any comments on this?

[alanruttenberg]

Two forums for discussion: https://groups.google.com/forum/#!forum/bfo-discuss, https://github.com/BFO-ontology/BFO-2020/issues I'm certainly interested in cases where this might be considered an unintended implication. Barry is updating the BFO 2.0 reference to document BFO-2020 - I'm not sure what the status of that effort is. The axioms, in a variety of formats, are at https://github.com/BFO-ontology/BFO-2020/tree/master/21838-2 The FOL is authoritative, and the OWL is specified as any axiom that can be derived from the FOL. The current contents of the OWL file are not complete wrt to the FOL (it's not a solved problem how to create the best approximation). Note that the first two issues reported by Werner Ceusters involve documentation and an error in some of the axioms we're discussing, on my agenda to fix in the near future. Issue reports are very much welcomed. An important function of this axiomatization is that it's something concrete to argue about.

Regarding the question of the implication of the dependents participating if the bearer does: Participation isn't particularly well defined, so we'll have to appeal to intuition. The use case that drove these axioms initially was that we wanted to talk about the specified outputs of data transformations. has_specified_output is a subproperty of participation, and the outputs of data transformations are intuitively GDCs, so we wanted to be able to express that. At the same time, BFO tries to ground everything in independent continuants, and of course, in any particular data transformation there's going to be some bearer, and it's quite reasonable to say that it participates as well. In most of the cases of data transformation we don't actually care what the bearer is, but that's not a problem. We don't have to say anything about it. There are other examples, like where the specified output of say, coloring something, is the color (SDC). Certainly the thing being colored participates by any intuition of what participation means.

The other direction does not have such obvious examples, and does have examples that don't seem intuitive. There are all sorts of qualities of molecules, but the qualities we would consider relevant in a reaction are more related to valence then to, say, the neutron count/isotope of the constituent atoms (first biological example that comes to mind). Because of this I'd be comfortable with an implication that charge participates, but a bit surprised if isotope participated. I'm sure we can come up with more examples. Because of this a general implication that dependents participate didn't seem justified.

Alan

On Thu, Oct 15, 2020 at 5:48 PM Nomi Harris [email protected] wrote:

@alanruttenberg https://github.com/alanruttenberg Any comments on this?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/oborel/obo-relations/issues/315#issuecomment-709608508, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAB3CDSYKS7E6ZQ6S32GLT3SK5U3ZANCNFSM4HDQJK2Q .

[nlharris]

Thanks, @alanruttenberg!

[cmungall]

I am closing this as the original question is satisfied - the range of has-input is continuant, so can be used for information etc

there are a lot of other unresolved issues in the discussions in this issue but someone will have to make new issues for them

[nlharris]

Chris said "I am closing this", but didn't close it. Closing now; reopen if needed (but preferably, create a new issue with action items).