[gmx-users] Re: FEP softcore problematics

[mernst at tricity.wsu.edu]

> Sorry about all of the confusion. Basically, the answer to this
> depends on what exactly you're doing. (I should really write all of
> this up in a tutorial soon). The way I think about it is that there
> are two fundamental principles you need to keep in mind, and you can
> do it any way you can cook up as long as you follow these two:
> (1) Always use soft core when modifying vdW interactions
> (2) Never leave charges on atoms for which you are turning off the vdW
> interactions
>
> The corollary is that you should generally not change charges and LJ
> interactions at the same time, but this is more for efficiency reasons
> than anything else.
>
> So, if you can cook up a way to do what you want that doesn't violate
> principles (1) or (2) and only involves two steps, great. For my
> purposes, I'm doing absolute binding free energy calculations where I
> don't mutate one thing into something else, so I can always do this --
> I'm just disappearing atoms, so I first turn off teh charges, and then
> second turn off the LJ interactions.
>
> More generally, I suppose if you can think of a way to do the
> transformation you want that only involves either (a) inserting atoms,
> or (b) deleting atoms, but not both, you can do it in two steps. If it
> only involves insertion, you first modify the LJ interactions to
> "appear" some dummy atoms, and then turn on their charges. If it only
> involves deletion, you first turn off the charges, then you turn off
> the LJ interactions. If it involves both insertions and deletions,
> maybe there is a way to do it in only two steps, but it isn't obvious
> to me.
Thank you very much for the clarification. None of my calculations have involved
combined insertion/deletion. I have always picked the state that has more atoms for my
starting state, then mutated toward the state with fewer atoms. This means that I don't
have to re-do recent simulations (hurray!) but I'm still not sure why some of them give
the results they do (boo!).

I am amazed what I have learned about free energy calculations from this list and from
papers suggested on it (Michael Shirts' recent work in particular, even if I don't need
as much accuracy and precision as he achieved). The documentation for most packages that
support free energy simulations is almost criminally simple in its description of the
process. When I first started investigating thermodynamic perturbation/integration, it
seemed like it would be fairly straightforward to use. I thought maybe it would increase
setup time by about 50% and simulation time tenfold. Now it seems like it's practically
a discipline unto itself. Review articles and original papers from 10-15 years ago also
made me think the process would be much simpler than it actually turns out to be. Even
books from the last 5 years haven't given me nearly so much useful information as
references and suggestions from this list have. There's a big gap between getting
roughly plausible results and getting reliable quantitative information, and it seems
that most published material I've seen never crosses that gap.

Again, thank you very much for your patient support and education that goes hand in hand
with your excellent simulation software.

Matt Ernst
Washington State University