Hi Thomas,
As I said I have not made systematic comparisons. Much of force field
development and validation is about philosophy of what the developer feels
is the best target. All of these ff94 variants (and charmm and others)
have a very limited functional form. They aren't going to work well in all
cases. I think it's important to understand what the goals in fitting were,
and to make sure that the properties used in fitting and validation are
similar to your project goals. These vary widely. It would be great if
every force field worked for all problems, but then we'd probably have
poorly converged data. We often sacrifice broad transferability for speed
(accuracy for precision).
I'm sure these issues aren't new to you - I'm mostly saying them so others
following this thread know my opinions since they do influence my choice of
targets for fitting and validation in the SB force fields. I'm aware that
we can't do everything well, so I do try to put effort into understanding
the limitations in the models I develop. We did some of that I'm the
original 99SB article, but it's hard to do systematic studies when so many
variants get developed. Those are valuable studies, though, so when people
see them it's very helpful to give them close study and see if the testing
matches the type of applications that you plan to carry out. In my
experience that's crucial, since there simply won't be a "best" force field
given the simple functional form.
This is all a long way of qualifying my previous statement. I would not say
that 14sb is the best force field. For what we do, it works well. We do
have studies that aren't out yet showing what I think is significant
improvement for the side chains. The backbone tweaks are fairly minor, as
they were in ff99sb* and other variants people made to our ff99sb backbone
parameters. It's likely some will work better for some systems than others
(IDPs for example). We use 14sb because it was designed to work well for
the kinds of things I do, and what some other people do.
I guess what I'm saying is that people should not expect there to be one
force field better than all others (yet). All include some character of the
people that designed them, and they probably work best in the domain where
they had training data. As I tell people here often, my suggested strategy
is to ask on the support lists how to solve a technical problem, but for
"best practices" it's always best to look in the literature for successful
studies that addresses problems similar to your own goals, and follow their
protocols. In other words, start with the science, and pick the methods
that work best. Don't start with an idea of the best method and assume it
will work for your goals.
Again, Thomas - I don't mean this to be a direct response to you since
you're quite experienced. This thread just made me think about voicing some
opinions that I have. I expect others will have different opinions - that's
why much of this field is still an art and we don't all use the same tools.
Best regards, Carlos
On Sep 12, 2014 8:40 PM, "Thomas Evangelidis" <tevang3.gmail.com> wrote:
> On 12 September 2014 16:43, Jason Swails <jason.swails.gmail.com> wrote:
>
> > On Fri, 2014-09-12 at 18:48 +0530, sunita.tifrh.res.in wrote:
> > > Dear Guys,
> > >
> > > After going through the literature I find ff99SB*-ILDN is the one which
> > > gives comparable result to the experimental observations.
> > >
> > > I am having Amber12 where leaprc of ff99SBildn, ff12SB etc are there
> but
> > > ff99SB*ildn is not there.
> >
> > I've never heard of this force field before. When I pulled up one of
> > D.E. Shaw's papers mentioning ff99SB*ILDN [1], that force field had 3
> > citations: ff99SB, another citation, and then the original citation for
> > ff99SB-ILDN (i.e., there is no indication I see that this force field is
> > any different than ff99SB-ILDN).
> >
> >
> I don't like making bold statements, but in this case you are totally
> wrong. The paper describing the backbone torsion modifications made in
> ff99SB and ff03 by Best & Hummer (as Hannes correctly mentioned) is:
>
> http://pubs.acs.org/doi/abs/10.1021/jp901540t
>
> According to Scopus this paper has 180 citations, so I am surprised you
> never heard of ff99sb* or ff03*. ff99sb*-ILDN is simply ff99sb* along with
> the side chain torsion modification introduced by Shaw and co-workers.
>
> In my experience ff99sb* along with charmm22* are the golden standards used
> by people that do metadynamics - and not only- with GROMACS. I tend to more
> and more people to use these "star" force fields in the last years and I
> bet there must be a good reason for that.
>
> As far as I am concerned, I do accelerated MD (in AMBER) and metadynamics
> (in GROMACS) of intrinsically disordered proteins and I can say that I am
> disappointed by the performance of ff12sb and ff99sb-nmr1-ildn. These seem
> to overestimate the helical content of the cases I am working on. In
> contrast, ff03w+TIP4P/2005 (another modified amber force field) which was
> designed to yield more accurate helix-coil transitions, gives better
> estimates of the secondary structure, at least for my cases. The latter
> test was made by HREX simulations in GROMACS, which is very computer
> demanding. I believe the addition of these modified amber force fields
> (ff99sb*-ildn, ff03*, ff03w+TIP4P/2005) would be a great upgrade to the
> AMBER arsenal. AMBER has the advantage to be faster in GPUs than GROMACS or
> NAMD and also provides aMD, which is one of the least computer demanding
> enhanced sampling methods. Unfortunately, I do not have the time, neither
> the technical support to port myself these force fields into AMBER.
>
> Now, regarding Carlos' statement that ff14sb is better than ff99sb*-ildn, I
> am curious to know the reasons that led him to that conclusion.
>
> best,
> Thomas
>
>
>
>
> --
>
> ======================================================================
>
> Thomas Evangelidis
>
> PhD student
> University of Athens
> Faculty of Pharmacy
> Department of Pharmaceutical Chemistry
> Panepistimioupoli-Zografou
> 157 71 Athens
> GREECE
>
> email: tevang.pharm.uoa.gr
>
> tevang3.gmail.com
>
>
> website: https://sites.google.com/site/thomasevangelidishomepage/
>
> ===============================================================
>
> *Physics is the only real science. The rest are just stamp collecting.*
>
> *- Ernest Rutherford*
> _______________________________________________
> 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 Fri Sep 12 2014 - 18:30:02 PDT
