Chronological Thread 

< Chronological >      < Thread >

---

The argument for versioning is simply the same kind of argument for
code versioning. If the goal is to represent the paper, then successive
versions should just converge on that. Using the 'HEAD' versions should
technically give you the most accurate model.

The questions of 'variants' is probably the more critical one. A very
simple example is taking a model that correctly represents a paper and
works and generalizing it so that we have a general model and an
instance model that imports it. One could say the instance model now
represents the same overall model as the original non generalized
model, but I would call this a variant and have metadata that provides
a semantically and formally defined concept about this variant as well
as reciprocal relations to what it is a variant of. Further instance
models that imported the generalized model may simply form a set of
experimental instances. This is only one case.

The proposal is that model relations are explicitly represented in the
metadata, which will allow for human readable(through the model
repository interface) and machine interpretable relations between
model.

The variables we have then to uniquely identify a model are 'name',
'verssion', and 'variant'. The version is the axis of least
variability.

For the Zhang et al. rabbit sino-atrial cell models, I would say the
multicellular version is a variant. The model that corrects errors in
the published model could be a version or a variant, I think that is
the point to debate. For instance the model that clearly represents the
publication, including its errors, then there is a case to say this
remains as it is and that any successive versions simply correct errors
in representing that publication. I would tend to put the second model
that fixes errors as a variant, because it does not represent the
publication anymore. However, some may argue that it better represents
the intention of the publication, especially if it now produces results
that match what were present in the publication. Whatever the answer,
it is important to formally represent the relations between variants
and perhaps version, which is what we are proposing.

cheers
Matt



On 26/06/2005, at 10:48 AM, David Nickerson wrote:

> Alan's right about the different versions, they should all be properly
> annotated with metadata to show whats different - which should be
> forced by the workflow involved adding models to the repository under
> the new system (as I understand it). Then a user of the model
> repository would be able to browse this metadata, including any
> validation information that we might add, and make an informed
> decision as to which version of a model they wish to include into
> their model/simulation/etc. For some of the popular models, we may
> also want to look at making recommendations as to which version we
> think best represents a given model, allowing people to easily use the
> best LRd or Noble '98 model without worrying about the differences in
> versions.
>
> As for the published models, I think we're still taking the position
> that we need to provide accurate representation of published models,
> including any published corrections. This is where multiple versions
> of a model come into play, so that people can access the model as
> published (maybe not too useful) or get the one that has been
> "corrected" and is capable of duplicating the results published in the
> original paper. Sometimes there will also be the case where multiple
> people have corrected the same model in slightly different ways, and
> we want to be able to make these differrent versions available in the
> repository.
>
> And then there is the whole issue of when a modification to an
> existing model becomes a new model or just a new version of the same
> model...
>
> Anything to add Matt and Poul ??
>
>
> Andre.
>
>
> penny.noble AT physiol.ox.ac.uk
> wrote:
>> To add a little bit to what Alan has said one thing that struck me is
>> that
>> the COR version is and should be the same as the original paper... the
>> only difference (and quite a crucial difference) being that all errors
>> from the paper have been removed... in other words it is a correctly
>> functioning model. Unless I am misunderstanding something here I
>> don't see
>> any reason to distinguish between those two versions?
>> Alan Garny said:
>>> Good news re the CellML website.
>>>
>>> As for having several versions of a CellML model, I have the feeling
>>> that
>>> it
>>> goes against the whole idea behind CellML and trying to make the
>>> exchange
>>> of
>>> models as easy as possible. Say that there are several versions of a
>>> model
>>> that you are interested in, which one to choose? One reason for
>>> someone to
>>> use a particular model is that s/he thinks it may be relevant for
>>> what
>>> s/he
>>> is doing, but that doesn't mean that s/he knows the details of that
>>> model
>>> and therefore is in a position to decide on which version of the
>>> model to
>>> go
>>> for.
>>>
>>> Maybe you meant that there would be several versions that are clearly
>>> documented. For instance, like for the Zhang et al. rabbit
>>> sino-atrial
>>> cell
>>> model that I sent to you guys. There are 3 versions, one that is the
>>> purely
>>> published model and two others, including one that is a corrected
>>> version
>>> of
>>> the published model and another for use in a multicellular settings.
>>> Having
>>> different versions for this kind of reason would, in my view, would
>>> make
>>> sense, rather than having different versions because of the software
>>> that
>>> was used to create/modify them.
>>>
>>> Alan.
>>>
>>>
>>>> -----Original Message-----
>>>> From: David Nickerson
>>>> [mailto:d.nickerson AT auckland.ac.nz]
>>>> Sent: 24 June 2005 23:57
>>>> To: Alan Garny
>>>> Cc: 'Penny Noble'; 'Peter Villiger'; 'Denis NOBLE'
>>>> Subject: Re: CellML saga...
>>>>
>>>> Thanks Alan (and Penny)!
>>>>
>>>> As you've probably noticed we have now switched
>>>> www.cellml.org over to the plone CMS system. Matt is just now
>>>> debugging his model repository plone/zope/python code and
>>>> then we'll be in a position to have a properly controlled
>>>> work flow for entering models into the CellML model
>>>> repository. This will hopefully allow for multiple versions
>>>> of a model in the repository (e.g., original paper, one coded
>>>> by Penny using COR, one modified slightly for use in CMISS, etc...).
>>>>
>>>> For now, I think Peter Villiger has been looking at splitting
>>>> the models up into separate components, illustrating how to
>>>> use CellML 1.1.
>>>>
>>>>
>>>> Thanks,
>>>> Andre.
>>>>
>>>>
>>>> Alan Garny wrote:
>>>>
>>>>> Hi David,
>>>>>
>>>>> Here are 2 more CellML files to be added to you guys' collection:
>>>>>
>>>>> - earm_noble_model_1990.cml: rabbit atrial cell model,
>>>>> - nygren_atrial_model_1998.cml: human atrial cell model.
>>>>>
>>>>> Cheers, Alan.
>>>>>
>>>>> --
>>>>> University of Oxford, Department of Physiology
>>>>> Parks Road, Oxford OX1 3PT, England Phone:
>>>>
>>>> +44 (0)1865 272-501
>>>>
>>>>> http://noble.physiol.ox.ac.uk/ Fax:
>>>>
>>>> +44 (0)1865 272-554
>>>

---

• [cellml-discussion] Re: CellML saga..., Matt Halstead, 06/26/2005
  • [cellml-discussion] Re: CellML saga..., Melanie Nelson, 06/27/2005

---

Archive powered by MHonArc 2.6.18.