Forums

--- Stefan Hoops <shoops@vbi.vt.edu> wrote:
> Darren Wilkinson <darrenjwilkinson@btinternet.com> wrote:
> > In both cases these are entirely comparable. What you call
> > "events per time" is essentially "molecules per time", and
> that
> > _is_ "substance per time".
>
> This is only true if the stoichiometry is 1 :), i.e., you need
> more
> information than the rate law to do the conversion.

I don't see why. It's exactly the same as the continuous
deterministic case - you multiply the reaction rate by the
stoichiometry to get the rate of substance transfer for a
particular species.

> I do not dispute that intelligently written software can take
> all this
> into account and convert from an ODE model to a stochastic
> model and/or
> vice versa. We are actually doing that in COPASI (ODE ->
> Stochastic).
> However, I do think that relying on intelligent programmers or
> scientist
> to do the conversion is not the way it should be. Having the
> option of
> one rate law for ODEs and another for stochastic simulation
> would be
> preferable in my opinion. One area of application could be
> hybrid
> algorithms which adapt the way a particular reaction is
> treated.

An alternative approach (which I favour) is to encode the entire
model for discrete stochastic simulation, and then leave it to
the simulator to figure out how to treat each species. After
all, there isn't simply discrete stochastic and continuous
deterministic. What about continuous stochastic, for example?
You don't really want to label that differently again, do you?

> They are not fundamentally different things.
> They
> > are exactly the same thing, measured in directly comparable
> > units. Treating them as different things when they are
> > essentially the same will lead to a needless divergence. One
> of
> > the strengths of SBML is that is captures the essentials of
> the
> > dynamics of a biochemical network without being explicitly
> > linked to a particular chemical kinetic theory. Let's not
> invent
> > a split where none exists.
>
> I think this is the only point were we are not completely on
> the
> same page. The continuous discussion and questions as well as
> the list
> of problems on the SBML site taught me that we can either live
> with the
> current kludge (intelligent software conversion) or invent
> something
> better.

Actually, we are probably on the same page even here. SBO
annotations are supposed to help with exactly this point.
However, my own view is that the SBML itself should be left
essentially as-is in this regard, as I think it is perfectly
adequate for capturing the essentials of reaction kinetics,
independently of the modelling framework. As you point out, we
don't want to restrict the development of more intelligent
simulation tools in the future by artificially categorising
models (or even reactions) at this stage as being either
continuous-deterministic or discrete-stochastic.

Cheers,


--
Darren Wilkinson
email: darrenjwilkinson@btinternet.com
home www: http://www.darrenjwilkinson.btinternet.co.uk/
work www: http://www.staff.ncl.ac.uk/d.j.wilkinson/