| Author | Topic |
|---|
Posts: 0
Registered:
September 2003
|
|
thoughts on models and and model parameters
|
05 Dec '02 06:55
|
 |
|
Just some thoughts on models and model parameters...
(1) It would be good, at some point not too far in the future, to be
able to represent (a) uncertainty in parameter values and (b) a known
distribution of values for a particular parameter. Had a post couple
months ago with references/examples.
(2) Unlike many other types of physical systems that have a hard layout
and connectivity map, biological systems seem likely to, in part, have
more of an "amorphous" architecture. One thing this might impact is
the entire notion of network and system layout. For example, one
Protein likely doesn't only interact with one to three other proteins,
but many many proteins across a range of interaction strengths (see
attached figure; also available here:
http://web.mit.edu/endy/www/scraps/KdDistro2.jpg).
So, for me at least, I don't see the model and its parameter values as
being separable at all. It may be a long term issue though (i.e.,
return to in 2010) and something that will be more relevant to protein
interaction-based networks versus metabolic networks et cetera.
best,
-drew
|
|
|
Posts: 123
Registered:
September 2003
|
|
Re: thoughts on models and and model parameters
|
07 Dec '02 14:43
|
|
|
> So, for me at least, I don't see the model and its parameter values as
> being separable at all. It may be a long term issue though (i.e.,
> return to in 2010) and something that will be more relevant to protein
> interaction-based networks versus metabolic networks et cetera.
Actually I also agree that parameters and models are not independent. This
is as true in metabolic networks as it is in protein binding networks. It
turns out that enzymatic activity depends upon substrate binding to
enzymes and this is also a spectrum of values (i.e. most dehydrogenases
can use NADPH and NADH with different affinities, though some are small
enough that we donīt usually consider the alternate substrate, but even
this may be relevant at times)
Although models may be incomplete as some parameter values are not known
(or uncertain), isolated parameter sets are completely meaningless. This
is because parameter sets only exist in the context of a model. So even if
one could have models with several parameter sets, there will never be a
parameter set that applies to two models. This means, from a computational
point of view, that making parameter sets independent from models is not
useful. So, we have arguments from both the biochemical and computational
sides to refuse this idea of separating parameter values from models.
Pedro,
looking forward to a lively workshop!
|
|
|
Posts: 55
Registered:
September 2003
|
|
Re: thoughts on models and and model parameters
|
08 Dec '02 02:07
|
|
|
Hi Pedro and Drew,
> > So, for me at least, I don't see the model and its parameter values as
> > being separable at all. It may be a long term issue though (i.e.,
> > return to in 2010) and something that will be more relevant to protein
> > interaction-based networks versus metabolic networks et cetera.
>
> Actually I also agree that parameters and models are not independent. This
> is as true in metabolic networks as it is in protein binding networks. It
> turns out that enzymatic activity depends upon substrate binding to
> enzymes and this is also a spectrum of values (i.e. most dehydrogenases
> can use NADPH and NADH with different affinities, though some are small
> enough that we don$B!-(Bt usually consider the alternate substrate, but even
> this may be relevant at times)
>
> Although models may be incomplete as some parameter values are not known
> (or uncertain), isolated parameter sets are completely meaningless. This
> is because parameter sets only exist in the context of a model. So even if
> one could have models with several parameter sets, there will never be a
> parameter set that applies to two models. This means, from a computational
> point of view, that making parameter sets independent from models is not
> useful. So, we have arguments from both the biochemical and computational
> sides to refuse this idea of separating parameter values from models.
I don't think so. For model *parameters*, ok, yes, because they can
have arbitrary semantics valid only in the model. But for *constants*,
no, at least in some cases. Given organism and measurement conditions
defined, many constants in rate equations can be shared by different
models. In this way we can have a efficient means of collaborating with
wet labs.
This is especially true for mesoscopic constants such as affinities and
small ks. Even if the model include non-orthogonal nature of
protein-ligands and protein-protein interaction networks, I think it
should be constructed ab initio from these fundamental constants, at
least from computational viewpoint.
My idea was that if we decide to provide a separate format for parameter
sets (or constants), it should be able to work as a means of
facilitating publishing and exchanging of measured data for various
organisms and measurement conditions.
But it is now depend on design decision of SBML itself; to define the
SBML document including the parameters, or to make it modular by
defining the dynamic model (as opposed to just a model) as a set of a
SBML document plus zero or more parameter set documents. Being a
high-level description of biological models (as opposed to
simulator-specific low-level ones such as VCML and EML) I still think
the latter is more elegant from software engineering viewpoint.
But this may need further discussions. Because it is a design decision,
I like the latter but I'd be happy with the former once decided.
regards,
Kouichi
|
|
|
Posts: 237
Registered:
September 2003
|
|
RE: thoughts on models and and model parameters
|
09 Dec '02 02:24
|
|
|
on Pedros comments:
I agree that:
>parameter sets only
>exist in the context of a model.
however this doesn't mean that more than one set of
parameters is meaningful for the same model. In fact
this becomes a very meaningful way to represent mutants.
An additional application (which might be close to Pedro's heart)
is using parameter sets to store the results of different
parameter optimisation/fitting runs on the same model.
[Whilst I can see that a modeller can use a database of
experimental results directly (across several models) my
impression is that
a significant part of the modeling community is not doing
this i.e. there is a significant amount of massaging,
optimisation and guessing going on that seperates
the raw experimental results from the parameters
used in models. However I don't know this from
direct experience. I'm interested in knowing more about
other peoples experiences.]
I think we can make parameter sets a seperate standard
that references sbml models. At worst parameter sets
would be in a rdf element(?) in a seperate XML namespace
in the same stream/file as the model. At best the
parameter set would be in a seperate file.
I think we should not encourage the use of parameter sets
in publishing contexts at least until the dust settles on this
issue with wide adoption in software.
I guess that all we need in SBML L2 is parameter value
and species initialAmount to be optional.
It is clear that different tools are going to have to
do there own validation of some aspects of SBML.
eg presence of parameter values.
However I would hope that parser libraries could
have a function like checkHasValues which raises
errors if parameter values and species initialAmounts are left out.
Later paraser libraries could do parameter value resolution from
a 'stack' of parameter sets that are applied to a given model.
yours Andrew
> -----Original Message-----
> From: owner-sbml-discuss@symbio.jst.go.jp
> [mailto:owner-sbml-discuss@symbio.jst.go.jp] On Behalf Of Pedro Mendes
> Sent: 07 December 2002 22:43
> To: sbml-discuss@symbio.jst.go.jp
> Subject: Re: [sbml-discuss] thoughts on models and and model
> parameters
>
>
> > So, for me at least, I don't see the model and its
> parameter values as
> > being separable at all. It may be a long term issue though (i.e.,
> > return to in 2010) and something that will be more relevant
> to protein
> > interaction-based networks versus metabolic networks et cetera.
>
> Actually I also agree that parameters and models are not
> independent. This is as true in metabolic networks as it is
> in protein binding networks. It turns out that enzymatic
> activity depends upon substrate binding to enzymes and this
> is also a spectrum of values (i.e. most dehydrogenases can
> use NADPH and NADH with different affinities, though some are
> small enough that we donīt usually consider the alternate
> substrate, but even this may be relevant at times)
>
> Although models may be incomplete as some parameter values
> are not known (or uncertain), isolated parameter sets are
> completely meaningless. This is because parameter sets only
> exist in the context of a model. So even if one could have
> models with several parameter sets, there will never be a
> parameter set that applies to two models. This means, from a
> computational point of view, that making parameter sets
> independent from models is not useful. So, we have arguments
> from both the biochemical and computational sides to refuse
> this idea of separating parameter values from models.
>
> Pedro,
> looking forward to a lively workshop!
>
>
|
|
|
Posts: 26
Registered:
September 2003
|
|
|
Posts: 237
Registered:
September 2003
|
|
|
