Alife and all that

[Oded Maler 920806]

I don't want to get into all that, just correct an apparent
misconception that some CSG-ers seem to have: behavior-based
robotics (a-la Brooks) is just one branch (not the main one)
in the Alife coalition. The mainstream is more related to things
such as cellular automata, self-replication, emergent proerties,
self-organization, complex dynamics and this kind of stuff.
I also think that the "crowd" demo is the PCT work which is closer
in the spirit to some Alife research.

Of course AL shares *some* of the stupidities of AI (but refrains from
others [and has its own special ones {but *every community* has its own,
isn't it?}]).

And in this spirit, since nobody objects to casual forwarding,
I enclose something I found today incidently in sci.chem,
concerning simulation of living cells.




From irisa!corton!mcsun!uknet!gdt!aber!hrs Thu Aug 6 13:06:12 1992

Article: 6487 of sci.chem
Xref: irisa sci.chem:6487
Path: irisa!corton!mcsun!uknet!gdt!aber!hrs
From: (Herbert Martin Sauro)
Subject: Simulating the Cell
Message-ID: <>
Date: 5 Aug 92 13:42:32 GMT
Organization: University of Wales, Aberystwyth
Lines: 92

I was interested to read that someone wants to simulate the chemical
reactions occuring inside a cell. As this is my field of interest I
thought I might make a few comments. As a number of people have already
(and correctly in my view) suggested, the short answer to this is no you
can't easily simulate a cell. Forget about the need for a big computer,
that's a minor problem compared to the lack of data, i.e real data on
the kinetics of the enzymes that make up your cellular model. As someone
said in their reply, people have been trying to simulate pathways for
*years*, even before the digital computer was invented, pathway
simulations were being done on analogue computers. The late David
Garfinkel was a pioneer in this field. His largest model must have
comprised of up to 400 reactions. Of course lots of other people
(including myself) have since tried to simulate bits of cells. I pretty
much abandoned the whole idea though for two reasons. One is that there
is so little relevant data on enzymes that the model one ends up with
bearly resembles the real thing, and secondly if you do end up with a
model that manages to simulate a cell or some part of a cell, you have
the problem of interpretation. These systems can get so complicated
(it's not like your simple pendulum or inclined plane as in physics)
that one doesn't have a clue for example why a particular metabolite
should be going up in concentration while another is going down, and I'm
just thinking here of simple steady state behaviour let alone
transients, periodic or chaotic behaviours. There is also the comment
that if the model is as complicated as the real thing then one might
aswell study the real thing (unless of course you just like simulating
things with a computer).

I suppose it really depends on what one wants to get out of the model.
If you are building a model which you hope will reproduce how a real
cell behaves that I think you're in trouble (for the already mentioned
reasons), of course go ahead and have a go but you'll probably see what
I mean after a while. The other reason why you might want to simulate
pathways is to try and understand basic pathway behaviour. In other
words what sorts of behaviour can pathways elicit. At the most basic
level pathways can be in one of three states, either equilibrium (not
very interesting), steady state (including unstable steady states) or
transients, i.e moving from one state to another. So first of all one
has to decide what state one is going to investigate. Secondly,
something others have mentioned, are you going to assume a homogeneous
system (i.e no concentration gradients and therefore no PDEs, just ODEs)
or a heterogeneous system. Clearly a full PDE model (even in only two
dimensions) requires a big fast computer. This still leaves the problem of
interpretation of the results. Until a few years ago there was little or
no formal theory which could describe metabolic pathways. Of course
there is enzyme kinetics (a formal theory of enzyme behaviour), but
trying to extend this to whole pathways (even to a two step pathway) is
a hopeless task. The equations are simply insoluble (ok a two step
pathway is soluble but the solution is about 3 pages of A4 long!). So
where does one go from here? Since I abondoned realistic simulations I
have turned to Metabolic Control Analysis (MCA) (or a similar theory called
Biochemical Systems theory). These may sound a bit wooly, but believe me
they are not. They have probably made more of a contribution to our
understanding of pathways than all the simulations that have gone before.

The essential goal of MCA is to be able to relate the genotype of an
organism to its phenotype. This is of course the classical objective in
genetics (I really mean genetics here not cloning) and two of the
originators (Kacser and Burns) were geneticists. The two other
originators (Heinrich and Rapoport) were biochemists. It would be
difficult to properly discuss MCA here (if there is any interest I'll
mail a proper summary to the net) as this mail is already too long but
just to give you an idea....

MCA first of all separates the parameters in the cell (or pathway) from
the variables in the cell. If one is just studying a pathway then the
parameters are usually taken to be the boundary conditions, i.e source
and sink pools, the concentration of enzymes and their respective
kinetic parameters. For the sake of argument we can treat the expression
level of the enzymes to be the 'genotype' of the system. The variables
of the pathway are the concentrations of intermediate metabolites and
the pathway flux. These are what we consider to be the 'phenotype'. The
question that MCA then attempts to answer is how do the enzymes (through
their concentrations and kinetic parameters) determine the levels of
metabolites and flux through the pathway and how are the metabolite
concentrations and flux affected by changes (for example) in enzyme
levels. MCA attempts to answer this question by tracing both
qualitatively and quantitatively all the cause and effect communication
routes through the pathway. In this way we are able to build up a
picture of how the pathway is working and what would happen if it were
perturbed, say by a mutation.

The moral of the story is, yes go ahead, have a go at simulating the
cell, you never know, you might come up with something nobody else
has thought of or discovered.

Herbert Sauro
Herbert Sauro e-mail:
Biological Sciences phone: +44 970 622353
Univesity College of Wales


[From Rick Marken (920806)]

Oded Maler (920806) says:

robotics (a-la Brooks) is just one branch (not the main one)
in the Alife coalition. The mainstream is more related to things
such as cellular automata, self-replication, emergent proerties,
self-organization, complex dynamics and this kind of stuff.

The term "behavior-based" does seem to give away the non-PCT related
goals of Brooks' robotics effort. But I am familiar with work done under
some of those other rubrics (particularly complex dynamics) and I have found
no one in any of those areas who is trying to deal with the fact of control
in living systems.

I hate to be a broken record about this (the fact that control theory is
about control) but what can I do; nobody in these (and other) fields of
the life sciences seems to listen to this point so the inclination is to
say it (and try to demonstrate it) over and over; after all, I'm a control
system controlling for (amomg other things) the perception that others have
heard me.

I also think that the "crowd" demo is the PCT work which is closer
in the spirit to some Alife research.

On the surface maybe; but is the spirit of Alife to explain how complex
appearing behavior results from the efforts of individual organisms
to CONTROL perceptual variables? Do they see twhat is most interesting (and
important) about the individuals in "crowd" is the inputs they are trying
to produce for themselves, and not the outputs that are so compellingly
obvious on the screen?

If so, then, indeed, Alife and PCT are doing the same thing -- and that would
be great.

I hope that you (Oded) or Eric can find the time to explain some ALife project
in detail and show how it relates to PCT.

Best regards




Richard S. Marken USMail: 10459 Holman Ave
The Aerospace Corporation Los Angeles, CA 90024
(310) 336-6214 (day)
(310) 474-0313 (evening)