Re: non uniform attempts distribution in constant pH MD

From: Brian Radak (brian.radak_at_gmail.com)
Date: Wed Jun 13 2018 - 10:27:59 CDT

Hi,

The default inherent pKa choice is the reference compound value, which is
of course uniform over residues of the same type. If some of those residues
are/are not of interest, you can also use "cphProposalWeight" with one or
more residue selections to assign a weight (these default to 1.0 and are
normalized at run time).

If I am understanding correctly, you did not change either of those
settings - is that correct? Nonetheless, with 200 attempts and 157
residues, I'm not entirely sure I should be shocked that some residues were
never selected and others were selected multiple times, since rejections
based entirely on inherent pKa values do not explicitly show up (maybe they
should?). We are also working on improving the random number generation for
those routines, which may not be as good as it could be.

Cheers,
Brian

On Wed, Jun 13, 2018 at 6:04 AM, Yasser Bruno Ruiz Blanco <
ybruizblanco_at_gmail.com> wrote:

> Hi all,
>
> I am running cpH MD and curiosly looking at report of the number of
> attempts per residue after finishing some short runs (~200 cycles with 157
> titratable residues = max number of attempts, 10 000 switching steps and
> 5000 standard simulation steps)
>
> I understand this report refers to actual switching attempt computing the
> alchemical transformations, and not just to the prior selection attempts
> that can be neglected by the intrinsic Ka algorithms
>
> Given I did not set any particular intrinsic Ka value for any of the
> residues in the protein, I would expect NO effect of the intrinsic Ka
> filter among residues of the same type. (i.e. I expect that all ASP are
> sampled evenly or at least tend to so...)
>
> Interestingly I am seen that the sampling after 200 cycles is almost
> concentrated in a few (2 or 3) of the 157 residues.
>
> Is this a behavior that has been observed before? Note, the sampling is
> not fixed among the same 2 or 3 residues, but it is a fact that within 200
> cycles the sampling that I got is extremely unhomogenuos within the same
> residue type. Which is something that I would not expect if the residues
> are selected randomly...
>
> I'll thanks your comments.
> Best,
> Y.
>
>
>

This archive was generated by hypermail 2.1.6 : Sat Dec 07 2019 - 23:19:54 CST