Apparently the second time I send the message also failed, so here goes
again!
Herbert Sauro
Caltech
"Mike Cantor" <mcantor at SMI.Stanford.EDU> wrote in message
news:<4.2.0.58.20001212135204.01bb9ec0 at smi>...
> Hi folks,
>>> Before I ask my questions, I want to quickly mention the motivation for
> this paper. Peter and I began an exploration this summer into the idea of
> storing kinetic information in Ecocyc so that people could automatically
> generate kinetic models from the knowledge-base. Our underlying mistaken
> assumption, though it took us a few months to figure it out, was that
> kinetics was like genomics -- one community of people was busy spewing out
> volumes of reliable data for public consumption, while another set of
> people was trying to intergrate all this information from big databases to
> build models. One of the main goals of writing this paper from my point
of
> view, besides describing the state of the art, is to educate the
> bioinformatics community about how the field of kinetic modeling really
works
>
Hi there. Your project to put kinetic data into your database is of interest
to us. I'm part of a group at Caltech attempting to put together an XML
standard (SBML - systems biology markup language) for representing
biochemical networks. We're just about to finalise the first draft sometime
in the next few weeks, so far we've quite a bit of interest and positive
feedback from potential software implementers and users. The other thing
we're doing is developing a software infrastructure standard to allow the
various simulation packages to communicate (and future simulators) with each
other.
Details at:
http://www.cds.caltech.edu/erato/
How are you going to allow other programs access to the database? SQL, an
API? At the moment we discovered that many of the genomic databases out
there are very closed systems and don't allow access other than through
direct user interaction via web pages. This is a great shame as there is a
huge amount of data out there but much of it is not easily accessible.
> 1. There have been many approaches to metabolic modeling. Kinetic
> simulation and MCA seam to be the most widely used, but I have also seen
> papers on the use of , e.g. Discrete event based models, Petrinets,
> Tendency Modeling, etc. What do you see particularly promising (or not
> promising) techniques for generating what types of insights about
metabolism.
>
Personally I don't think you can do much better than good old fashioned ODE
representations. The other approaches you mention are tempting (ie
Petrinets, etc) but I wonder if any one can cite an actual instance where
these alternative methods have actually produced useful insight. Now I'm not
saying don't use them, of course all research avenues should be explored but
perhaps be aware of their limitations. Some approaches have proved useful,
such as network analysis particular when there is
limited data, eg only the network structure is available.
> 2. What are the prospects for larger-scale modeling projects like the
> NIGMS's projected 25$ glue grant for a "Virtual Cell." Is this a naive
and
> premature waste of money, or will it be a fountain of knowledge.
>
I don't know but I suspect not very good unless executed carefully. I was at
a conference in Tokyo some weeks ago and one of the talks went on about a
large cell model of some bacterium, hundreds of reactions. One of the things
they observed in their large model was a counter-intuitive movement in ATP
levels upon Glucose starvation (an observation that apparently was worthy of
a Science news report) - ATP went up in the initial phase of starvation
before declining (A biochemist looking at a metabolic chart will quickly
realise why this might happen). The surprise to me was to be told that it
took them *6 weeks* to work out why this happened. The problem was possibly
that their model was so complex that they simply couldn't understand what
was going on - it should be pointed out that they did not appear to use any
analysis tool other than time course simulations (which certainly didn't
help), now the use of even a moderate amount of MCA or some similar analysis
may well have helped them much more.
The take home message for me was two fold, large models may not be such a
good way to approach cellular modelling and secondly the selective use of
analysis tools other than time course simulation is probably very important
and I'm sure it would have cut down their time to work out what was going
on. The use of common sense is also a very powerful weapon.
There is one approach that large model advocates should perhaps apply and
that is model reduction. Start with the large model (if you must) and then
reduce it by taking out those steps etc which do not appear to contribute to
the behaviour that one is interested in.
> 3. What is the relation between metabolic modeling and informatics? What
> if anything can the field of metabolic modeling gain from the genomics
> explosion?
>
Probably one thing that metabolic modelling could gain from genomics is
network structure information and possibly crude interaction data but others
may have other
ideas on this.
> 4. What is really the role of all of the metabolic simulation software
> packages I see out there, such as Gepasi, SCAMP, Mist, DBSolve, Kinsim,
> SCOP, etc. Are these widely used? What factors have led to their
adoption?
For me, exploratory tools. They help me think about the problem, a
laboratory for testing hypotheses, new equations etc. Once one has explored
hypothetical models (simple ones preferably) you can apply your experience
to more realistic models, very useful I think, probably if the people who
built the large bacterial model I described above had done this they
wouldn't have taken 6 weeks to work out why ATP went up upon Glucose
starvation. Building a realistic model of a heart, liver etc is not such a
big pull for me, I know a lot of people however like to do this.
I think you'll find that many of these simulation packages were written by
people in order to aid them in their own personal research, only after, did
they emerge from the lab and get used by others.
Simply based on a troll through the citation index, you'll find that the
majority of metabolic models have been done on Gepasi and Scamp, roughly
50/50. Only recently has computational biology got the attention of the
grant awarders and the popular scientific press and so interesting times are
probably ahead.
Herbert Sauro
Caltech
>> Thanks so much to anyone who replies to this posting. I have really
> enjoying reading the work of many of the subscribers of this list.
>> Cheers,
> Mike Cantor
>>
