[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: 'not an artist' article

On 05-Jul-2000 Steve Baker wrote:
> Erik wrote:


> The critical point where it would all (hypothetically) fall apart - even
> presuming
> we had the compute power to do it - is that we currently have no good
> algorithms
> to predict how a protein will fold into a 3D structure given only it's
> chemical
> formula.
> Since much of protein chemistry is largely determined by key-in-socket type
> mechanical contact - we are essentially screwed.  Knowing the list of A's,
> G's, T's and C's will (with a lot of work) get you the chemical formulae
> of the proteins - but there is no way to figure out what those proteins
> actually *do* in the resulting critter unless you have prior knowledge
> that a protein found with such-and-such formula is known (experimentally)
> to cause some specific effect in the creature in nature.

but we're looking for a much higher level result. :) Take thermodynamics for
example... There are things like kinetic theory that say why the pressure
increases if temperature increases in an isochoric PvT system. It's useful to
know the kinetic theory, and it's relevent to thermodynsmics, but it's not
actually used... it's just used to establish the simpler rules like PV=nRT or
(whatever Van der Waals gas is). We can observe the low level approach that
genome project lays out to develop the simpler approaches we'll use.

> I believe that there is a prize offered to anyone who can write a program
> to predict protein folding in sensible amounts of time - it's a huge prize -
> maybe a million dollars.  If you can do this - you are in the wrong business!
> In any case, the number of genes for a real critter is way more than 50. For
> a human, they expect to ultimately discover somewhere between 20,000 and
> 500,000 depending on who you listen to.  They have only found like 10,000
> of them - and of those, there are only a few hundred that they know the
> functions of.

some company a few months back patented a new method and my understaning was
very recently the decoded it all, there was gonna be some presidential speech
and all that jazz. 

But even if they haven't found all the genes, we don't want to use the actual
gene, just observe how they work together, and find a suitable model. By taking
all the genetic theory and constructing a model that is easy to compute yet
still relatively accurate, we can generate convincing creatures. And since
we're applying the genetic algorithm to find a likely set instead of an optimal
solution, we can skip trying to group optimal components, and simply create
random merges (possibly with a heavy mutation factor).

> For a program like biomorphs, a couple of dozen genes is all we could
> really afford to implement.  That's enough to describe a 2D monochrome
> fractal - but THAT is enough to draw pictures of trees, planes, insects,
> etc.

> Anyway, this is all rather a red herring - we could be talking about
> evolving convincing polygon meshes for bicycles...the fact that we are
> using a genetic algorithm to create the triangle mesh we want -
> which happens to represent a living creature that evolved using a
> genetic algorithm - should be regarded as a coincidence.

such an algorithm can be applied to almost anything. My suggestion of using
actual data was so we can accurately describe the 'problem'. If you're using a
GA to find a near-optimal solution to a traveling salesman problem, you'll need
to know certain things in regard to geometry, like calculating the distance
between points, etc. I think it's all a matter of describing the problem :)

> We don't have 50 million years to wait - and we don't care whether
> we have the details of anapurinol synthesis in bone marrow exactly
> right for our 8 foot tall purple mutant penguinoid.
>   :-)

that kinda detail would be pointless :) Overkill, even. But we can see how the
real thing works and build our model in a similar fashion. Possibly using a
subset of the real data.


        -Erik <erik@smluc.org> [http://math.smsu.edu/~br0ke]

The opinions expressed by me are not necessarily opinions. In all
probability, they are random rambling, and to be ignored. Failure to ignore
may result in severe boredom or confusion. Shake well before opening. Keep

To unsubscribe, e-mail: linuxgames-unsubscribe@sunsite.auc.dk
For additional commands, e-mail: linuxgames-help@sunsite.auc.dk