From: Peter Freddolino (
Date: Tue Aug 22 2006 - 22:54:31 CDT

Just to follow up on John's post:
the new ionization tool we're going to be posting is designed for use
without solvent. It's about the simplest ion placement algorithm you can
get, really; it calculates coulombic energies due to the protein on a
grid around the protein, and then starts placing ions in the minima,
updating the grid after each placement. This isn't necessarily an ideal
tool (which would probably be an iterative poisson-boltzmann solver),
but it is simple, effective, and easy to parallelize on any processor
set that shares memory. Running on, say, 24 cpus on an SGI altix (like
ncsa cobalt) lets you ionize a 160,000 atom structure (which doesn't
count solvent -- this is the STMV structure) in about 30 minutes.
Hopefully this is of interest to you. We'll post usage instructions
along with the binary to the MDTools website in the next couple days.

John Stone wrote:
> Michel,
> On Tue, Aug 22, 2006 at 09:18:06PM -0500, L. Michel Espinoza-Fonseca wrote:
>> Hi John,
>> Thanks for helping me with this. I've done what you suggested and yes,
>> when I load both psf and pdb files into VMD, they almost used most of
>> the RAM I have available. I did it only running VMD with no graphical
>> interface and no other program that can use that memory.
> I loaded up the structure and looked at it in the "points" representation
> for speed, and colored it by position, index, and a few other things.
> Very cool :-)
>> It if fine
>> while I'm not running autoionize. Then the requirement or memory goes
>> up to 3.6 GB and after a while it crashes.
> Yeah, I think that you're going to want to use a different ion placement
> tool on this actually, as the autoionize plugin is going to need more memory,
> and it'll take a long time to run on the 8 million atoms with all of
> that water. Some of my colleagues and I have written another ion placement
> tool that is probably better suited to your problem than autoionize is,
> but still need to to package it up and post it. I'll ask Peter to post
> about it. The main advantages are that it's all C/C++, runs in parallel
> on shared memory machines using pthreads, and should work much better for
> huge structures like you're working with.
>> Maybe we can try the version of VMD that you're going to release,
>> although I can wait for a couple of days if it coutd get you into some
>> troubles. Hopefully this time it'll work out.
> Yeah, after 1.8.5 is released I'll see what I can do to make you a
> "lightweight" build of VMD that needs a less memory per atom.
> John