发信人: heculase (似水年华), 信区: Biology
标 题: Re: 推荐一些比较有意思的细胞的实验室或program吧
发信站: BBS 未名空间站 (Mon Sep 12 22:05:48 2005), 转信

Good post....

Agree with your idea that biological systems are generally robust and
got few determining parameters upon thousands of variables. Low
dimensional chaos seldom exist, in fact they can exist but determining
parameters are controlled to somewhere mapping is single-valued.
Nobody only interested in mathematical structure of symbolic dynamical
system, though it does tell something about possible bifurcation etc.

To respond to the point of massive randomness, a counter example will be
from genetics, where 10 or more alleles on duplicated sites together
determine a phenotype, combination makes the phenotype map almost
continous. Randomness here does play important role in selection and
drift. If this is a gene on butterfly pattern determination, we got
continuous or chaotic patterns. However it is a unbalanced example to
hippo's statement -- since for butterfly the selection pressure for
reproduction, during speciation, is much larger than the one for
survival probability to reproduction day.

To speak reversely, hippo's point is correct because there are thousands
of other genes but only these few, with hundreds or so alleles on them,
totally determined the fate of a population.

Hence here comes the question about quantative measurement: What are
those factors, which factor is important, and how we determine them
among thousands.

An important contribution from mathematical dynamics is that
biological feasible attractors are always semicontinuous, which implies
considerably lower dimensional kernel (as hippo point out -- very
good point here) -- and emergence of such a kernel attractor from
collective activities. If you do not mind, this is the phaseshift from
oscillation to grandmother cell. Topological structure and theoretical
analysis helps here to figure out the important nodes in networks, where I do
think modeling will play a key role. Not only finding these nodes
after the whole network is solved out, but to point out where lies the
possible important node well before a full picture is drawn -- this may be
exciting.

Nevertheless the design principles, in genomics, neural network and
ecological systems, are largely -- if not totally -- different. Though
large populations of random variables exist in all three, they generally
different in behavior. Here I do not initiate mathematical argument,
but I do not think if we can do something on genomics then it implies we
can do something on neural networks, vice versa. I am a bit more
optimistic on neural network since the interaction is relatively simpler
than gene networks, and activities could be tracked down much more easily.

However I am still a bit pessmetic to model magnetism from spins.

A reason should be there's very few tools for spatiotemporal dynamics besides
statistics and pattern analysis. However these are not at all fundamental.

【 在 puppeteer (减肥中的河马) 的大作中提到: 】
: I think the fact is that very few systems in biology are low-dimensional
chaos
: as far as I know. Typically there are a lot of parameters.
: I think in biology, it is really the opposite, we have a high dimensional
: system with a lot of unreliable components, it's easy to get randomness. The

: challenge is the system is actually very robust and sometimes controlled by
a
: few parameters because it has to respond to a more or less predictable
: environment. This is why I think there is hope in the modeling approach.
: Ultimately maybe we will learn to speak the language without knowing all the

: detailed parameters. There might be organizing rules which are the design
: principles of these systems.
: ...................

--
CSBi@M, SysBio@H, HST@M&H, CDS@Caltech, CNL@Salk
All for one and one for alL.

"the university cares alot about us getting skin cancer so they put us
in the basement" Christof Koch



※ 修改:·heculase 于 Sep 13 00:31:51 修改本文·[FROM: 144.82.]
※ 来源:·BBS 未名空间站 mitbbs.com·[FROM: 144.82.]