Re: 'not an artist' article

[Erik <br0ke@math.smsu.edu>]

On 05-Jul-2000 Pierre Phaneuf wrote:
> Erik wrote:
> 
>> I'd imagine that quite a few of the 'simpler' have very limited
>> sets of genes, and fairly controlled patterns. Take atomic assembly
>> at the first state of excitation, there is a very simple and definite
>> way that atoms are put together, and given just 2 numbers you can
>> accurately model a stable atom. You can also easily model interactions
>> between atoms that create molecules. There're a lot of different atoms
>> (check a periodic table), and a HUGE number of molecules. But with a
>> small amount of 'control' data you can accurately construct atoms
>> and molecules.
> 
> I am in the high performance supercomputing area (supporting NEC SX-4
> and SX-5 vector machines, one of them with 128 gigabytes of RAM), and my
> girlfriend is a microbiologist doing a master related to molecular
> biology (genes and proteins).
> 
> This is so far away from being possible, you can't possibly have an
> idea.
> 

you also misinterpret what I meant, and didn't see my reply ...

> Proteins are a chain of amino acids tied together to form a single long
> molecule (a polymer). DNA is transcribed into RNA, which then goes into
> a ribosome to be translated into a protein by assembling the smaller
> molecules. Proteins are NOT straight, they are twisted all over
> themselves because of various atomic forces that attract and push away
> other atoms. The shape of the resulting protein is called its
> conformation, and it determines what the protein will actually do, by
> having various "activation sites" at specific places in its structure.
> For example, the hemoglobin protein is shaped like a ring, with iron
> atoms along the inside part of the ring, which binds with oxygen atoms
> to carry them.
> 

My suggestion did not involve modeling this, but simply understanding how this
works together and making a more advanced conceptual layer, which can be built
on to generate a plausible algorithm.

> Determining what amino acids will be produced by the translation process
> is easy. Determining their conformation (which is what determine their
> function, remember) is amazingly hard. Even the largest and most
> powerful supercomputers can barely predict how two or three amino acids
> will conform, and there are hundreds or even thousands of amino acids in
> proteins!
> 

And the models we use should NOT attempt to address the amino acid reactions.

> When this will be possible, we will have attained a very important
> milestone of molecular biology. Currently, we find out what genes do by
> trial and error, passing them through ribosomes and checking out what
> protein comes out of it, but when this problem will be solved,
> scientists will be able to say "okay, I'd need to create a protein with
> this and this feature" and have a computer tell them what string of DNA
> will given them what they want. It is the difference between finding
> something and using it as well as you can to do what you want, and
> designing something specifically made to solve the problem at hand.
> 

I'm not looking to further molecular biology, I'm looking to take what IS
known, hack it into a plausible model, and fake it.

> In the meantime, we suck. ;-)
> 

that's the kind of attitude that hampers development of this region. Going back
to thermodynamics, they understood molecules, they understood physics, but
their models were NOT originally based on them. They took what they know of the
lower level stuff, and built an abstract model from that and from observation.
On a physics level, I can describe what happens to an object released in a
gravity well based on all sorts of nasty particle computations with lots of
integrals describing the object and the center of mass, and describe how it
falls at a molecular level. Or I can describe how it falls as an object, using
a couple incredibly simple formulas. While it is important to understand the
molecular level, that's *NOT* the one that gets used. A character generation
engine should *NOT* attempt to use the genes to assemble amino acids and
generate a character from that. It SHOULD take the information known about
genetics, and create a very simple yet mostly accurate set of formulas that can
be implemented easily and effeciently. I would be annoyed if such an algorithm
took more than an hour on a 486.

I *DO* think an understanding of real life genetics should be understood. I do
*NOT* think they should be applied directly at the lowest level.

I'm not sure, but maybe a better example would have been generating a simple
tree or linked list in software. A solid understanding of how memory is handled
at an opcode level, (allocated, accessed) should be known, and the methods were
developed with those things in mind. However, when I implement a linked list or
tree, I'm not hacking it out in assembly, I'm using a higher level approach that
summarizes large chunks of the process in very simple small instructions. I
understand page tables, segments, offsets, etc... but I let that be abstracted
with 'malloc'. If I didn't have a fairly good knowlege of pointers and low level
memory handling, I would probably have problems with dynamically allocated data
structures. Even though we are able to understand the concepts for a very low
level, we don't work there. We just keep it in mind and generate the algorithms
at a higher level.

I think by understanding some fundamentals of how the genes work together and
what kind of things they address, we can generate a higher level approach to
'faked' genetics using a very small set of (primarily) artificial genes with
some pretty direct policies (again, possibly artificial). As the more generic
and abstract models and gene-sets are experimented with, they can be adjust to
closer model the nasty details of the real thing. We may be able to use some
real genes in the mix, possibly not. It was never my intention to indicate that
we should simply take all this raw data from the genome projects, cram it into
our POS x86's, and expect biomolecularly perfect results in acceptable times.
I've done enough big number projects to know that the home computer is incable
of a LOT of interesting things. But we get around limitations like that by
faking it where we can. I've yet to see a 3d game (or renderer) with 'perfect'
rendering... Tricks like raytracing, raycasting, radiocity, polar rendering,
etc are used to 'fake' the creation of a visual image.

        -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
Refrigerated.
        

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