Nicholas Smith

A survey to determine functional requirements for each domain

Questions


  • What type of CellML models would you / do you normally work with?
Cardiac Electrophysiology

  • Have you used any other CellML tools before? If so, which ones? What did you like and dislike about each one? What about graphing and analysis tools?
mozCellML: Was slow and crashed occasionally. Liked the ability to integrate, and the ability to modify things in versions. On the right general track.
LabHeart: Liked stimulus protocols, also the interface, the speed, and that it was specific to electrophysiology.
COR: Liked the mathematics editor.
  • Do you usually work with a single model at a time, or do you compare multiple models to each other?
Mainly work with a single model, but occasionally want to compare them.
  • Could you describe some typical scenarios for how you work with models (i.e. the steps you take when working with CellML models).
  1. Checking that a model does what you think, by comparing with the paper. This involves finding a steady state, and comparing with experimental data.
  2. Combine models, by taking components and assembling them to form a new model.
  3. Optimising parameters, and then putting those parameters back into the original model.
  4. Evaluating stiffness of models (eigenvalues) then changing integrators.
  5. Trying to reduce models, adding new constraints.
  6. Checking for thermodynamic consistency.
  • Do the models you work with have many initial conditions, or just a few? Can you think of a rule for determining which initial conditions are useful? Would it make sense for model authors to provide this information, or does it have to be set on a model-by-model basis?
There is a clear distinction between parameters (values that can be changed) and initial conditions (values which shouldn't be changed), and the model author should describe what are parameters and what are initial conditions.

  • Are you happy editing initial conditions as a floating point number, or do you prefer some other representation?
Happy editing as a floating point number.
  • Do you tend to regularly change all the initial conditions in your models, or do you tend to just change a few, and find that the others are irrelevant to individual problems?
Only a few at a time. It would be useful to hide others, but this would need to be specified while working with the model.
  • Do you often run a model multiple times with different initial conditions? If so, what sort of analyses would you want to do that take data from both runs?
Yes. Initially plot graphs and look at these.
  • When plotting graphs, do you find that you only plot a limited number of graph types, or do you end up plotting just about everything against everything? Would you find an interface which automatically detected variable types from the units or metadata, and produced certain types of graphs more useful than an interface which let you plot anything against anything?
There are a limited number of graphs, but a generic interface would still be more useful. It would, however, be useful for the model author to produce figures from the paper.

  • Can you list some graph types you use (for example, what is on each axis, and how the variables can be identified). When several dimensions (e.g. ampere and ampere per square metre) may be used depending on the model, please list them all and/or explain the rule.
Not applicable.

  • What sort of changes to the models do you make when you are using (as opposed to writing) the model? Do changes to initial conditions cover it, or do you need to change certain parts of the model? Can you think of a way to identify these expressions ahead of time?
Only during creation, except for the ability to specify stimulus protocol (piecewise linear, piecewise quads and cubics).

  • Would a template-driven model designer (for example, which allows you to put together components which perform certain tasks from a list, and automatically assemble them into a working model) be useful to you? If so, can you give an example of how it would be useful, and describe the sorts of parameter tweaking you do to put such models together.

Not at this stage, but perhaps in the future, because models not currently robust enough for this to work.

  • Do you ever need to fit data(e.g. to recover model parameters from experimental data)? If so, do you always perform a certain type of fitting problem, or would a more general user interface suit you more?
Yes. A more general user interface would be the best.

  • Are there any sorts of diagrams used in your field which you would find useful while using the models?
Schematic of models with compartments and membranes. Viewing it is more important than being able to click on it.