Unsolved or Poorly Solved Computational Problems

Richard Scott rtscott at forgetspamPacbell.net
Wed Mar 5 19:20:10 EST 2003


Thank you for your thoughtful response. Some of the nomenclature (e.g.,
"ODE" and "DAE solver") that you use is unfamiliar to me because of my
relative ignorance.  However, the concrete examples, terms and references
that you have provided will be a real help to me.

I can't respond to every point of yours because there is more work to do
first, and because it might be fruitful for us to consider including
contributors from a different newsgroup.  Namely, those in the
"bionet.software.www" group that also responded to my posting there.  It was
unclear to me at the time which group was more suitable and my message was
posted there several days before it appeared here. I would really appreciate
any suggestions about how to combine these two conversations because both
seem to provide a rich divergence of viewpoints even among clear experts in
the field.  This may reflect the interdisciplinary nature of the subject
(biology, chemistry, math, computer science and so on) but I suspect that
the field is actually fragmented.

The most interesting points that you made are about the need for "dynamic"
modeling flexibility in regards to the pace of simulation time and method. I
have actually worked on combining wide ranges of time intervals in one
environment and switching simulation methods programmatically so that aspect
is not so challenging.

I need to learn more about "large biochemical networks".  If you know of any
additional keyword search terms or actual educational references please let
me know about them.  I am familiar with traditional engineering control
concepts but not biological control mechanisms and concepts.

I was surprised by your comment, "Straight forward simulation of a
biochemical is essentially a 'solved' problem and there is lots of software
available to this sort of work, so what we don't need is yet another
differential equation or stochastic system solver!".  In the short time that
I been searching the Web, I have not found dynamic models of two or more
interacting biological molecular chains.  The closest was the "MC" program
developed by UCSD's Baker and Holst which seems to be a simulation of the
electrostatics of complex chained biological molecular structure (
http://www.sciencenews.org/20010901/fob8ref.asp and
http://www.scicomp.ucsd.edu/~mholst/ ).  However, based on the experience of
the computer industry, I readily agree that going beyond stochastic models
seems very necessary. In that regard, I quote myself from the other
newsgroup, "This leads to two other issues which I have encountered in both
the academic
and commercial aspects of computer science, and which I suspect are big
problems in biotechnology. First, the lack of the right sort of experimental
data in key areas (e.g., actual disk accessing patterns on request by
request basis). Second, an unwillingness to design and construct tediously
accurate and complete simulation models (e.g., timing accurate disk
simulations).  The computer industry has no excuses for not properly
instrumenting key behavioral components in hardware but I imagine that a
similar effort in molecular chemistry or biology is virtually impossible
with today's technology. If that assumption is correct, there must be a
crucial need for accurate simulation capabilities to test various
theories.".

Your comments about "important (not just interesting) problems" are a useful
guide along with those of Kevin Karplus' and Bob Burner's.  If you combine
those comments you are led into issues of both problem definition and
perspective which is another full discussion.  Maybe next time after I
absorb some of this material.

Thank you again for your help,
Richard
"Caltech News Server" <hsauro at cxx.calxxx.edu> wrote in message
news:b3kfcb$88r$1 at naig.caltech.edu...
> There are lots of unsolved problems in computational biology.
>
> Here are a few:
>
> How can we combine stochastic and continuous modeling, even better, can we
> develop an algorithm that will switch parts of the dynamical model
> automatically from stochastic to ODEs as necessary?
>
> In almost all biochemical network simulations there will be fast and slow
> reactions, can we develop a DAE solver that will automatically partition
the
> reactions for us at runtime into fast and slow dynamics?
>
> What sort of theory do we need to understand very large biochemical
> networks? What are the computational issues in trying to simulate very
large
> networks?
>
> How can we apply classical control engineering to biological control
> systems, there is already a control theory of biochemical networks but how
> does this relate to engineering control theory?
>
> and the list goes on.
>
> Straight forward simulation of a biochemical is essentially a 'solved'
> problem and there is lots of software available to this sort of work, so
> what we don't need is yet another differential equation or stochastic
system
> solver!
>
> >What are the important (not just interesting) problems?
>
> My experience has been that interesting problems almost always invariable
> turn out to be extremely important, I would concentrate on interesting
> problems, that way you're more likely to make a significant and lasting
> contribution. Examples of this abound in biology, most of our ability to
> manipulate and sequence genomes arose as a result of work done in the 50s
> studying viruses which infected bacteria, very interesting, but arguably,
a
> complete waste of time as a piece of 'useful research'. But ultimately it
> led to one of the greatest revolutions in biology.
>
> If you're concerned with human health and welfare, I don't think this sort
> of work will make a significant contribution for many years, and even then
> it might only affect a small number of people. If however, you managed to
> persuade people to stop smoking then you'd save over 400,000 lives a year
in
> the USA alone!
>
> Perhaps the one disease class which might benefit from a computational
> approach is cancer, the statistics for 1985 were 340,000 deaths (excluding
> lung cancer). The sort of contribution that comp bio could make is
> understanding the cancer state from a control point of view. All cancers
are
> the result of a failure in the signaling control networks.
>
> Herbert Sauro
>
> "Richard Scott" <rtscott at forgetspamPacbell.net> wrote in message
> news:Vyy6a.1551$3S5.173279277 at newssvr15.news.prodigy.com...
> > Greetings,
> >
> > At work, I have been working on a high performance C++ library not based
> on
> > STL.  The applications area was not bioinformatics but I have recently
> > gotten interested in algorithmic development for chemical and biological
> > simulation, and sequence identification.  I have an engineering physics
> > background but very little knowledge of either biology or chemistry.
> >
> > What are the major unsolved or poorly solved computational problems in
> > computational biology that could potentially lead to real improvements
in
> > people's health and welfare?  I know this is a naive question but my
> > knowledge is rudimentary, and I don't know where to find a description
of
> > the problems or the potential solutions.
> >
> > I have been surfing the Web to find answers but some issues are unclear.
> > For example, is the sequencing problem solved to everyone's
satisfaction?
> > What kinds of models and algorithmic approaches are being used for
protein
> > folding?  What are the important (not just interesting) problems?
> >
> > Richard Scott
> >
>
>






More information about the Comp-bio mailing list