Well, my definition is essentially what you say about being mathematically identical that is the user can swap both codes and expect the "same" result (ignoring for the moment the different setup and the fact that sander can't do the lambda end states).
Of course, pmemd does dual-topology, as well as single-topology (in effect at least) for direct conversion of atoms through linear scaling. So it is a hybrid. The elimination of terms for pmemd is actually an artefact of the setup with leap. But that's because you may explicitly need to copy all or part of your system to get leap do the right thing (proteins). If you do (non-covalently bound) ligands you don't need this step obviously.
I don't quite understand your second paragraph about computing the averages.
I am not sure if there is currently sufficient evidence to say that MBAR is more efficient than TI or vice versa.
Cheers,
Hannes.
________________________________________
From: Jason Swails [jason.swails.gmail.com]
Sent: 28 January 2015 20:04
To: AMBER Mailing List
Subject: Re: [AMBER] FEP
On Wed, Jan 28, 2015 at 2:44 PM, <hannes.loeffler.stfc.ac.uk> wrote:
> Pmemd14 does TI in the same way as sander does.
Define "same way" :). There are important differences between how the
codes work (which is why the setups for both programs are so radically
different). The way I see it, pmemd DOES use a dual-topology approach...
you just have to crush those two topologies together and eliminate
duplicate terms beforehand.
Simplified a bit, sander computes a lambda-weighted average of the two
topology files by computing both endpoints in their entirety and averaging
everything. By contrast, pmemd computes the same lambda-weighted average
by averaging the parts that are *different* and adding it to the parts that
are not. While the implementation (and by extension, using that
implementation) is drastically different, the two approaches are
mathematically identical.
> Both codes provide MBAR data. So one can easily compare how old a hat
> TI really is :-)
>
>
Yea -- MBAR here is used to optimize the use of TI data. I don't see how
MBAR is a viable alternative to any TI protocol that can't be done as a
simple single-step annihilation with no intermediate lambda values.
All the best,
Jason
--
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Jan 28 2015 - 13:00:03 PST
