[gmx-users] Re: TI, sampling, sc_power, and sc_alpha

David Mobley dmobley at gmail.com
Tue May 30 19:12:06 CEST 2006


Matt,

Re sc-power, you *do* want sc-power=1. I am not sure about defaults.

>David, when you talk of splitting the vdW and charge
> mutations into separate stages, you mean initially (for example) setting the B-states to
> have only altered atom types (and hence vdW radii), then later extending the
> trajectories generated in each window while using a new .top file that also has altered
> charges in the B-states -- right?

Yes, more or less, although I prefer to run these in parallel rather
than in serial. Also, there is one big caveat: You should never (at
any stage) have a charged molecule with zero vdW radius, because
hydrogens in many water molecules have vdW radii of zero. If you have
a negatively charged point with a zero vdW radius, the hydrogens can
overlap with this point and (obviously) stick to it extremely
strongly. (Also,  it's worth noting that this same thing could happen
if you change the charges and the vdw parameters at the same time,
depending on the transformation you are doing).

You can avoid this in various ways. For example, if you are only
modifying vdW radii, this probably isn't a problem. If you are
disappearing atoms, you should turn off their charges before turning
off the vdW interactions.

I'm doing absolute binding free energy calculations, which involve
disappearing ligands from binding sites. I typically do this by (a)
first restraining the ligand to stay in the binding site (see Boresch
2003) (b) running one set of calculations to turn off the ligand
electrostatic interactions while leaving the atom types fixed, and (c)
in parallel run another set of calculations to turn off the ligand van
der waals interactions where electrostatic interactions are already
turned off in both A and B states.

The obvious advantage of running these in parallel is
parallelizability. But I also don't like running the charge/vdw
calculations back to back for several other reasons:
(a) If you decide later that you need to extend a calculation to
reduce the uncertainty in that component, you have to run everything
again from scratch
(b) I think it introduces the potential for hysteresis
(c) If you have properly equilibrated your system at each lambda value
you shouldn't need to do this.

Others may disagree on this point.

Best wishes,
David


> Again, thank you for your help in handling this tricky topic.
>
> Matt Ernst
> Washington State University
>
> _______________________________________________
> gmx-users mailing list    gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.
> Can't post? Read http://www.gromacs.org/mailing_lists/users.php
>



More information about the gromacs.org_gmx-users mailing list