Replica exchange probability value

[casalini...@gmail.com]

Dear all,

I am currently using Gromacs 5.0.2 patched with PLUMED 2.0 in order to study the folding of a small protein (16 residues).

I adopted PTWTE approach, initially biasing potential energy in order to improve the energy overlap between replicas.

I have one curiosity: which values of exchange probability can be considered acceptable and meaningful?

In other words: should I add hills so that my average exchange probability between replicas is about 20% ... or about 70% ?

I thank you all in advance for your help.

Kind regards,

Tommaso Casalini

[peter.rich...@gmail.com]

Hey Tommaso,

Liu et al. 2005:
http://www.pnas.org/content/102/39/13749.full
uses 22 - 29 %

Wang et al. 2011:
http://pubs.acs.org/doi/abs/10.1021/jp204407d
20-30 %

Moors et al. 2011:
http://pubs.acs.org/doi/abs/10.1021/ct100493v
uses 20 %

Bussi et al.
http://www.tandfonline.com/doi/abs/10.1080/00268976.2013.824126
uses 35 - 50 %

Something in that range gives you a homogenous climb along the replica ladder.
Since you are using Gromacs you find the exchange probability printed at the end of the md0.log file.

An additional question from my side:
If you have a system with 3 replicas and only one exchange. If 1 exchanges with 2 or 2 exchanges with 3 gives you a probability of 100 %. Is that correct?

Cheers
Peter

[Giovanni Bussi]

Hi,

Concerning the simulation with an odd number of replicas: I think gromacs doesn't accumulate the average when the exchange is not attempted. Otherwise, you would always see an apparent higher acceptance for first two and last two replicas.

Concerning the right acceptance value: I don't think there is a unique rule. With N replicas you need more or less (N*N/acceptance) attempts to make a round trip, but attempts should be uncorrelated for this formula to hold. Whereas this is often true with T-REMD, be careful since in other H-REMD replicas get stuck more easily (this is a heuristic observation). So, check with demux.pl if replicas are actually traveling up and down.

Regards,

Giovanni

 

--
You received this message because you are subscribed to the Google Groups "PLUMED users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to plumed-users...@googlegroups.com.
To post to this group, send email to plumed...@googlegroups.com.
Visit this group at http://groups.google.com/group/plumed-users.
To view this discussion on the web visit https://groups.google.com/d/msgid/plumed-users/efaeced3-842c-4933-9209-1564f1075b93%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[peter.rich...@gmail.com]

Am Donnerstag, 21. Mai 2015 16:56:18 UTC+2 schrieb Giovanni Bussi:

Hi,

Concerning the simulation with an odd number of replicas: I think gromacs doesn't accumulate the average when the exchange is not attempted. Otherwise, you would always see an apparent higher acceptance for first two and last two replicas.

For a 9 replica system I get in the md0.log file in Gromacs:

Replica exchange statistics
Repl  5500 attempts, 2750 odd, 2750 even
Repl  average probabilities:
Repl     0    1    2    3    4    5    6    7    8
Repl      .54  .53  .56  .56  .55  .55  .57  .56
Repl  number of exchanges:
Repl     0    1    2    3    4    5    6    7    8
Repl     1483 1460 1519 1520 1492 1488 1528 1525
Repl  average number of exchanges:
Repl     0    1    2    3    4    5    6    7    8
Repl      .54  .53  .55  .55  .54  .54  .56  .55

Consider exchange 1 with 2: 5500 attempts, 1460 times successful -> 0.265 probability. Why is the value in Gromacs exactly doubled? Is that because neighbours do not exchange. if 1-2 exchanges, 0-1 and 2-3 do not exchange?

Is that what you mean by  "gromacs doesn't accumulate for not attempted exchanges?

[Giovanni Bussi]

Yes I think that is ok. But you have doubts perhaps you should discuss this on the gromacs mailing list.

Giovanni

To view this discussion on the web visit https://groups.google.com/d/msgid/plumed-users/27b96e3e-a6ea-40d4-acaa-77032233761c%40googlegroups.com.

[Carlo Camilloni]

2750 odd means that 1 <-> 2 has been tried 2750 times.

To view this discussion on the web visit https://groups.google.com/d/msgid/plumed-users/CAPLm8ZLOqdnRXD-PhdEu0DfAxpbe4x2dGTETQSQVJVVJ%3Dew3%2Bw%40mail.gmail.com.

[peter.rich...@gmail.com]

Ahh, now it is clear:

The replica_index.xvg file (produced by demux.pl) has 5500 lines = 5500 attempts, as set by the -replex option in mdrun.
The first attempt checks only:
0 <-> 1, 2 <-> 3, 4 <-> 5, ... the even exchanges

The second attempt (second line in replica_index.xvg) checks only:
1 <-> 2, 3 <-> 4, 5 <-> 6, ... the odd exchanges
and do not try the rest. That gives you the correct probability values in the .log file.

Thanks to all!
All the best!
Peter