Chronological Thread 

< Chronological >      < Thread >

---

Quoting David Nickerson <d.nickerson at auckland.ac.nz>:

> Just been thinking about how we could test/specify that models in the
> CellML model repository are MIRIAM-compliant. For a model to be
> "MIRIAM-compliant" it must pass all the tests (box 3) and contain all
> required annotations (box 4) specified in the MIRIAM article
> (http://www.nature.com/nbt/journal/v23/n12/abs/nbt1156.html).
>
> The six tests can be summarised as:
>
> 1. The model must be encoded in a public, machine readable format.
> 2. The encoded model must comply with the standard in which it is encoded.
> 3. The model must be clearly related to a single reference description
> that describes or references a set of results that one can expect to
> reproduce using the model.
> 4. The encoded model must reflect the biological processes listed in the
> reference description.
> 5. The encoded model must be instantiated in a simulation (including:
> the values of quantitative variables and their units must be equivalent
> to the values listed in the reference description).
> 6. The model, when instantiated within a suitable simulation
> environment, must be able to reproduce all relevant results given in the
> reference description.
>
> Test 1 is easy since all models in the repository are in CellML and test
> 2 can be automated using a CellML validator.
>
> Test 3 can be satisfied if we could somehow indicate the reference
> description for a given model - currently these are generally specified
> using a <bqs:reference> to the journal publication so we just need to
> tag or reference the approriate piece of metadata as the reference
> description for the model. This gets more complicated for CellML 1.1
> models that split a single journal publication into multiple models or
> combine multiple journal articles into one model.
>
> I can't think of a way to automatically perform test 4, so it probably
> requires new metadata in a model that specifies the result of this test,
> added either by the model author (preferably) or model curator.

Clearly this depends on the format of the reference description. If the
reference description is a journal article written in a natural language, I
don't think we can realistically expect to support this.

If the reference description is written in some machine-readable metadata
language, on the other hand, we can easily write software to perform this
check.

>
> Tests 5 and 6 could be peformed automatically if simulation results have
> been included with a model in the repository but we probably still need
> to manually specify that the model parameter values and initial
> conditions etc. match those given in the reference description (at least
> until the CellML models start becoming the reference description).

I think we can immediately claim that if there are too few initial values, the
model is not MIRIAM compliant(although authors are free to separate the
initial
conditions into a separate file which imports the mathematics).

Due to the changes to support equations without an analytic solution,
mozCellML
will now detect both over-constrained and under-constrained models(and lacking
initial conditions results in an under-constrained model error). Therefore, if
the model loads and runs in mozCellML without producing errors or not-a-number
results, it can be assumed to be MIRIAM compliant. However, there are be valid
CellML models which do not run in mozCellML, e.g. those that use <matrix/> or
<vector/>, but which may be MIRIAM compliant.

Test 6 requires reference data in a standardised format(CSV?) but when this is
provided(perhaps from a URI provided by the author, or converted into the
appropriate form by the curator) it can be automated.

> We also want to start including metadata specifing what simulation
> environments have been used with each model and what kind of success has
> been had, so checks for tests 5 and 6 could be implied from this data.
>
> Model annotation required for MIRIAM-compliance:
>
> 1. The preferred name of the model.
> 2. A citation of the reference description with which the model is
> associated.
> 3. Name and contact information for the model creators, that is, the
> people who actually contributed to the encoding of the model in its
> present form.
> 4. The date and time of creation, and the date and time of last
> modification.
> 5. A precise statement about the terms of distribution.
>
> The preferred name of a model is given by the model element's name
> attribute, right? Which will be appropriately generated/set on model
> upload to the repository.
>
> As with test 3, requirement 2 can be satisfied once we can indicate the
> reference description for a model.
>
> Requirement 3 is satisfied if a model contains the model creator
> metadata from the section 4.1 of the CellML Metadata specification and
> modification's can be included through the modification history metadata
> (section 4.6).
>
> Even the example in the MIRIAM paper doesn't include the actual time of
> creation/modifcation, but requirement 4 is satisfied through the use of
> the creation date metadata (section 4.5) and modification history
> metadata includes the date of modifications.
>
> The CellML metadata specification provides for specification of
> copyright notices (section 4.4) but we probably need to extend this to
> cover the range required by the 5th annotation requirement, especially
> to enable the specification that a model is public vs confidental.
Dublin core has a license term, which is recommended to be a URI to the
license.
This allows users to determine if they recognise the URI as that of an
acceptable license, and use it. This still meets the requirements of MIRIAM,
provided that the license at the URL is a precise statement of the terms of
distribution(I don't think it would be appropriate for us to judge whether a
license agreement is sufficiently precise, so we might need to just pass
validation if a valid license URL is provided).

Best regards,
Andrew


----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.

---

• [cellml-discussion] MIRIAM compliance for the CellML model repository, David Nickerson, 02/01/2006
  • [cellml-discussion] MIRIAM compliance for the CellML model repository, Andrew Miller, 02/01/2006
    • [cellml-discussion] freezing, Peter Villiger, 02/03/2006

---

Archive powered by MHonArc 2.6.18.