Newsgroups: bionet.metabolic-reg
Subject: Re: Universal method
References: <CEwCt1.289 at festival.ed.ac.uk>
Henrik Kacser writes:
> I think what John (Woods) means when he says that he "is not convinced
> that a straightforward implementation would be feasable" he probably
> means that there is no automatic and foolproof method where you just
> press buttons.
That's pretty much what I mean, although I'm also concerned about that
even totally defined `buttons' may be impractical.
A travelling salesman has to visit a number of towns, but may visit
them in any order. He naturally prefers to visit them in the order
which will involve least travelling. Half a dozen appointments
present no problem --- he just digs out his copy of AutoRoute Express
(TM). To deal with more cities he writes a computer program which
will calculate all the possible distances and record the shortest.
This method is both automatic and foolproof. It is easily stated, and
indeed the computer program is trivial. But by the time we get to 10
cities, 3.6 million permutations exist. 100 cities take us to 10^158,
and 1000 take us the already unimaginably huge 10^2567. (The world's
only unbreakable cryptography system, RSA public key, works in exactly
this way. `Everybody' knows what must be done to break it, but a code
created in 10 minutes on a PC might take millions of supercomputer
hours to crack.)
>As to his championing mutation/selection as the method of choice, he
>should really aqauint himself with the literature on selection limits
>(or perhaps talk to a population geneticists and breeders). What kind of
>mutation would he envisage that produces a 25 fold (2500%) increase in
>the enzyme activity of each and every step of say the tryp pathway?
Form:
Coming from a genetic rather than biochemical background, I would say
that these are the MCA/T papers I understand best. My critique of the
`Universal method' was murky and handwaving for good reason ---
intended more as a discussion-starter than as a in-the-literature
response. The general tone of electronic discussion tends to be
somewhat less intense than the scientific literature --- I would have
studied the paper an awful lot harder if I had intended a formal
criticism. What `expertise' I have tends to lean heavily towards
then computational rather than the biochemical, especially at the
moment. Hence I was a little taken aback to see Rankin taking me to
task for `dismissing it lightly' --- and I apologize for any offence!
Substance:
The travelling salesman, and other optimization problems, can now be
effectively `solved' for large n. If this were not the case, most of
us would be working on far less powerful computers --- optimal
placement of components on a chip is effectively the travelling
salesman problem writ large (might optimal flux improvement be, as
well?).
Two current exciting methods exist for difficult minimization:
1) Simulated Annealing (SA) and its derivatives (such as Very Fast
Simulated Reannealing). These attempt to mock the process whereby
crystals form, effortlessly reaching the lowest possible thermodynamic
state. A simulated temperature steadily decreases, whilst random
alterations to the existing solution are invoked if (a) the current
temperatue is sufficient to overcome the energy barrier and (b) the
change is favourable. ("I realize that technically that's just one
condition, but it was so important I thought I'd mention it twice"
--- Kryton, Red Dwarf). The slower the temperature cools, the longer
the optimization takes, and the more likely it is to settle into a
`good' result. Think of a ball bouncing on a pitted surface,
becoming less and less likely to overcome the barriers to other
regions.
2) Genetic Algorithms (GA) and derivatives work on the principle of
natural selection, with the `fittest' solutions surviving at the
expense of suboptimal ones. They are comparable in result to
evolution --- very good at complex optimization, but prone to fall
into deep local minima, with overwhelming change necessary for further
improvements. Just like the mamalian eye --- it's inside out because
the chance of inverting it without severe visual disruption is
minute, but it's now very well adapted for working inside out.
> What kind of mutation would he envisage that produces a 25 fold
> (2500%) increase in the enzyme activity of each and every step of say
> the tryp pathway?
I envisage such an increase being possible only with human and
computational intervention. The molecular biologists could use a
computerized version of the Universal Method and intervene directly as
indicated by the results. But I wouldn't rule out the possibility
of the computational approach using a GA (ie: successive rounds of
selection and mutation). This was all that my claim contended, albeit
stated with somewhat suboptimal clarity. I also beleive that VFSR is a
more likely source of fruitful solutions, but think it may have
generated rather less discussion.
>Watch this space.
>HK.
I think it's safe to say that we all are!
... John Woods
--
"
Forsan et haec olim meminisse iuvabit (Virgil)
[approx] Tr: The time may come when we will look back on these days and laugh
"
