Mike's minutes on CellML best practice / template meeting, Wednesday 2008-07-30

Mike's minutes on CellML best practice / template meeting, Wednesday 2008-07-30



Here are my recollections. Feel free to suggest additions/changes.


Enclosed is Cell Elite, run it by double clicking on the *.jar file. If your system doesn’t allow this, you need to get a recent Java Virtual Machine installation, probably best to see the IT people.





James calls meeting querying whether the way we do 'best practice' models is the same and fits in with Sarala's tool.


Discussion prompted by Catherine and Mike about how what they consider to be entities are also called

'processes' in Sarala's system. Poul made point that a differential equation for an entity is a conservation

process. Mike queried about model abstractions that are deliberately not conservation (ie species held constant

due to abstracting away some aspect of metabolism). Poul makes point that it could be a conservation with some

inputs being negligable or zero. Mike agrees this makes the model easy to extend if those previously abstracted-away

metabolic aspects are then modelled.


Query from Mike about whether the repository should be of CellML models or CellMLOWL models. Poul says that it CellML

models can easily be converted to CellMLOWL models and that it doesn't matter which is used.


Sarala and Mike agreed that the ways they consider 'best practice' do actually coincide with each other.


Discussion of use of template cellml components prompted by Poul. All agreed it would be a good idea. Mike mentions his

work with Synthetic biology which has yielded some templates. Poul asks if they exist as components and Mike says

they're in flat 1.0 format right now as without good tools it's painful to do. But conceptually they are there.

Mike mentioned has cut up the Pandit model (electrophysiology) into 'atomic'

components. James queries whether Mike used Andre's "decompose" tool, Mike says no as it uses encapsulation which is

not desirable.


Discussion initiated on the merits of encapsulation as currently implemented. Main points:


                - encapsulation hides component structure, variables and mathematics, and of those one the last is the one

                                that we can be sure we want to hide. Component structure may need to change to accomodate model

                                overlap, and variables of any kind might be useful depending on what else is connected (scaffold

                                protein modelling for ion channels given as example).

                - even if we don't use encapsulation, we still need a way of partitioning up models to import a whole bunch

                                in one hit with few connections. 'containment' is closer but is being phased out and

                                has physical containment implications.

                - higher level functions, while composed of accesible lower-level atomic components, should be importable

                  without importing explicitly all 'sub-components'.


                ==> We probably need a 'virtual containment' relationship in cellml for this new kind of modelling.


                - 1.1 modelling from atomic components means a lot of connection drawing, crying out for some visual tool

                                to do this. With all of Mikes 1.1 models for example, it is simple to see what to do but it takes a week or

                                messing around in a txt editor to implement it.

                ==> The possibility of a full-time tool person was mentioned.

                ==> Mike to send 'Cell Elite' to James and Catherine