X-Message-Number: 12943
From: Eugene Leitl <>
Date: Tue, 14 Dec 1999 12:36:49 -0800 (PST)
Subject: more on big computers and the protein folding problem
References: <>

CryoNet writes:
 > From: Thomas Donaldson <>
 > Subject: more on big computers and the protein folding problem
 > Date: Tue, 14 Dec 1999 01:29:38 +1100 (EST)
 > 
 > Hi again!
 > 
 > About protein folding: it's simply not true that we need a computer like
 > the kind that IBM is making to make progress in understanding protein 
 > folding --- though with current methods it will certainly help. There's
 > been a lot of successful work already aimed at understanding this problem;
 > what the big parallel computer will let us do is to work out folding for
 > much larger proteins.
 
I disagree. Imho, for a _robust all purpose_ protein folder you need
primarily two things 1) an accurate forcefield 2) enough computational
power to brute-force the problem by stepping through the whole MD
(molecular dynamics) trajectory. Why? Because the kinetics (and much
less thermodynamics) of the process determine the target energetic
pocket. It might or might not be the global energy minimum, but the
protein has no more clue about it than a mainframe, provided they both 
sample the local conformational space to about the same extent.
 
Current forcefields are not accurate enough by far, which can be
experimentally proven by taking a high-resolution structure from the
Brookhaven protein data bank (several 10 "benchmark" structures are
available), distorting it randomly, minimally, and attempting to fold
back using a current forcefield. In the vast majority of cases you
won't come back. So we also need a brand new TrueBlue forcefield to
boot.

 > It's even possible that someone will come along with an algorithm that
 > doesn't need such computer power, but such an algorithm (other than very
 > approximate methods which can refined by experiment) doesn't yet exist.
 
You might take a shortcut skipping the mouth of the folding funnel,
but the more you try to shortcut later, the higher probability to
misfold. Which reminds, certain proteins absolutely need chaperonins
to massage them into shape.

 > I know that most people on Cryonet don't bother to keep up with this side
 > of bioscience. I will say, though, that I have several times discussed

Well, not all of them.

 > new work on protein folding in PERIASTRON. A good number of clever ideas
 > have already been developed to study this problem; if anyone wishes they
 > can send me a message and I'll give them some references.
 > 
 > In fact, I personally suspect that IBM is building this computer not
 > because IBM people believe that it's needed to solve the protein folding
 > problem, but because previous work on protein folding has gotten far
 > enough that experts on the problem can now simply sit down and write
 > out the required programs. It's not needed to study the general problem
 > but rather to use what we already know to work out larger proteins which
 > cannot be worked out with a smaller computer.
 
Excuse me? If you can't solve the problem with current codes on
current iron, but can solve it using a much bigger machine, then you
effectively need a much bigger machine. This is not more, (but
certainly also no less), what Blue Gene is all about. Plus it gives
good PR.

 > After all, if you haven't noticed, for what it's worth, biochemistry 
 > now provides the ONLY fully developed "nanotechnology". And I put that

That's true, but this naturally evolved wet nanotech is currently
computationally infeasible to control. Notice that Blue Gene might be
also just powerful enough to brute-force the inverse protein folding
problem, which will be _highly_ useful both in medicine and for
drextech bootstrap.

Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=12943