Andrew Miller wrote:

>>   
> Have you actually looked at what exactly is happening? Is this an 
> overflow where quantities are getting so big they go to infinity because 
> you have a loop which doesn't have any self-regulation?
> 
Hi Andrew,

Yes, I have looked at what's happening. In some cases the integrator
only gets up to the second time point, which isn't that useful. For the
models that get further than that, I don't see any evidence of numbers
getting absurdly small or absurdly big. Although some of the fluxes in
the asthagiri_lauffenburger csv file that I've sent you seem to start
off extremely small - like 10E^-38 small. I don't know. That's with the
BDF integrator. With most of the other integrators (example Implicit
Gear M1) they get up to the second time step and then start producing
NaN outputs.]

> I presume that the model that is being generated from the reactions by 
> my script would be exactly identical to what you would enter if you 
> recreated the model by hand (assuming that the entered rate laws are an 
> accurate representation of the paper). If the rates were specified using 
> one of the common formulae for enzyme rates, you would get product 
> inhibition and these overflows wouldn't happen, but I think that perhaps 
> the entered rate laws are too simple.

I would presume that too.

> 
> Obviously, getting rid of the reactions is the overall goal, but I don't 
> see any reason why you would be better to re-create the models from 
> scratch, rather than starting from the original model with reactions, 

I am starting from the original model with reactions, not building it
totally from scratch.

> running it through my script to generate equations for this, and then 
> fixing any problems such as rate laws which don't represent what the 
> authors of the paper actually used and/or result in species getting 
> exponential increases to such high concentrations they overflow the 
> floating point representation.
> 

I don't think that this process is possible to automate due to the way
the models have been coded up using reaction components. Of course, I
may be wrong. It's something that's difficult to explain over email. If
you want to see what I mean, come see me and I'll show you an example.
Then perhaps you might have a better idea about whether the process can
be automated or now. Additionally, rebuilding the model provides a
really good opportunity for curation. I've been rebuilding the
bertram_arnot_zamponi model and I've actually found a few errors and
omissions, which I definitely wouldn't have found without looking at the
model in the depth required to rebuild it.

>>   
> My script could generate metadata as well if required. However, as Poul 
> suggested at the CellML meeting, feedback from Matt on exactly what 
> metadata we should generate it and where that metadata should be stored 
> (I think it preferably should be in the model, but Matt might have a 
> different opinion).

Yep.

Kind regards,
James
_______________________________________________
cellml-discussion mailing list
cellml-discussion@cellml.org
http://www.cellml.org/mailman/listinfo/cellml-discussion

Reply via email to