[AMBER] Specification of solventCap causing atoms to collapse into each other

From: Ben Roberts <roberts.qtp.ufl.edu>
Date: Tue, 23 Mar 2010 14:42:44 -0400

Hi all,

I've been trying to use XMIN to minimise a metalloprotein, and get a rather troubling result. I'm hoping someone can advise me on what needs to be done.

I started out with my protein in explicit solvent, and used ptraj to remove all solvent molecules and counter-ions, except those within 30 of a certain atom in the metal cluster; I'm using that atom as an origin for the solvent sphere. Writing the resulting molecule out as a PDB, I then used LeAP (specifically, sleap) to generate an appropriate inpcrd and prmtop. So far, so good.

I then ran 100 steps of XMIN minimisation, using igb=1, no periodic boundary conditions, etc., etc.

Using the solvent sphere described above with no modification, I get EGB = about -30,000 kcal/mol, a restraint energy of 0 kcal/mol, and the solvent water molecules have a tendency to boil off into space as the structure "relaxes".

If I specify that a solvent cap is used (in LeAP: solvateCap model TIP3PBOX <centre_atom> 30), only one water molecule is added - unsurprising, considering that the sphere is mostly full of water molecules already. Nevertheless, some important changes must have taken place, because I notice three new phenomena in the output, two of which I'm rather alarmed by.

First, the restraint energy is no longer zero, at least after the first few steps. I guess this behaviour is expected.

Second, the EGB increases from about -30,000 kcal/mol to +9,600,000 kcal/mol. (At the zeroth step, bonding components of the energy are equal, while non-bonding components are equal to within 1 or 2 kcal/mol, and the discrepancy can perhaps be explained the presence of one more water molecule.) I did not expect this behaviour; I would have thought that the EGB energy should be roughly the same whether a solventCap has been specified or not.

Third, and most worrying, sander's attempt to minimise the energy (which, by now, comes largely from the GB term) involves collapsing atoms in the metal cluster into each other, so that, after enough steps, five atoms share the same two points in space. Even after only 100 steps, two atoms have collapsed into each other, and another two are well on the way to doing the same.

This is not a matter of the bond and angle parameters; the bond and angle contributions to the energy rapidly increase as atoms run towards each other and the system becomes horribly distorted. Nor, as far as I can tell, is it a matter for the van der Waals or electrostatic parameters, as the atoms that collapse into each other are in 1-2 or 1-3 relationships anyway; but even if this were not the case, if the problem lay purely in the well-known non-bonding parameters (charge, sigma, epsilon) for the affected atoms, I should expect it to appear also in the simulation where the solvent cap is not applied, and I don't see it there.

Can anyone advise me on what the solvent cap is doing to cause this kind of bizarre behaviour from the generalised-Born potential, and how I could work around this problem in such a way that my solvent doesn't boil off into space (i.e., applying the restraints used by the solvent cap, manually if necessary, as I'm not sure what these are by default)?


Ben Roberts
AMBER mailing list
Received on Tue Mar 23 2010 - 12:00:04 PDT
Custom Search