Re: Protein RMSF in replica exchange simulations

From: Brian Radak (
Date: Tue Jan 26 2021 - 12:01:44 CST

In this case you can compute RMSF in the sorted trajectories and there
is no issue other than the usual sampling problems. A WHAM scheme may
reduce error but requires quite a bit more work.

On 1/25/21 5:26 PM, Victor Zhao wrote:
> Dear Brian,
> It is possible to rewrite the trajectories so that you have
> trajectories by replica rung. The sortreplicas script included with
> NAMD does that. I am asking about RMSF computed over a single rung,
> which would be made up of many individual trajectories as they passed
> through that rung.
> Best regards,
> Victor
>> On Jan 25, 2021, at 5:22 PM, Brian Radak <
>> <>> wrote:
>> Hi Victor,
>> How you calculate any observable (including RMSF) with replica
>> exchange depends on the implementation. I'm pretty sure (although I
>> might be wrong) that all of the REMD scripts distributed with NAMD
>> track replicas (i.e. the state parameter, here the REST2
>> "temperature", changes over the course of the trajectory).
>> In this case, the RMSF should be the same over all trajectories
>> within statistical uncertainty. This also means that:
>>  1. The trajectory statistics can be pooled under the assumption that
>> they are independent
>>  2. Any differences across trajectories are entirely sampling artifacts.
>> I would also contend that the RMSF computed this way is a completely
>> unphysical quantity that only gives you an impression of sampling
>> efficiency. It does not even describe real population thermodynamics.
>> For that you would have to apply a WHAM-like scheme. I don't know if
>> any scripts for that purpose are readily available.
>> HTH,
>> BKR
>> On 1/25/21 4:50 PM, Victor Zhao wrote:
>>> Dear all,
>>> I have been running replica exchange simulations (REST2). So far, I
>>> have focused on observables calculated on a per-frame basis and
>>> averaged to obtain ensemble values. But I am wondering, is it useful
>>> to calculate RMSF in a replica exchange simulation? Presumably you
>>> would do it over the replica index of interest (i.e. replica 0). But
>>> the trajectory at each replica setting includes exchanges between
>>> replicas. And yet I’ve read papers where RMSF is still calculated in
>>> replica exchange simulations.
>>> Often, RMSF calculations presented in papers don’t appear to make a
>>> larger point beyond that different parts of the protein differ in
>>> flexibility, so I think RMSF calculations in a replica exchange
>>> simulation could serve the same purpose. But when it comes to the
>>> issue of comparing RMSF between different replica exchange
>>> simulations, wouldn’t different exchange acceptance rates prevent
>>> comparison between simulations? That is, suppose one simulation has
>>> a 0 <-> 1 exchange rate of 0.2 whereas another has an exchange rate
>>> of 0.25. In this case, the average time between exchanges would be
>>> different, which could affect RMSF. I am wondering what the experts
>>> on this list think.
>>> Best regards,
>>> Victor Zhao
>>> PhD Candidate, Chemistry
>>> Shakhnovich Lab, Harvard University
>>> <>

This archive was generated by hypermail 2.1.6 : Fri Dec 31 2021 - 23:17:10 CST