Stefane,
why are you hesitant to reduce the size of your system. If you can
still address the same questions with a smaller system, then I'd say
the answer is nearly always to use that smaller system (REMD or not).
Besides, my gut feeling is that even if you had enough 271*16 cpus
available, you would not get enough exchanges within a reasonable wall
clock time to get a converged answer. I say this because you're
basically looking at a protein folding problem over 12*7 residues
(larger than I have seen REMD be successfully reported) but also you
have a big entropy problem because the 7-mers are not tethered
together so this might even be harder than protein folding. Sure,
there could be some help from symmetry, but I assume that you don't
know that yet or else why do REMD?
I am personally highly suspect about REST: what ensemble does this
method sample, exactly?
Implicit solvent might be an option, but only if your implicit solvent
aggregation energy landscape has important minima in the same place as
the explicit solvent energy landscape, and how would you know this?
So I suggest that you make your box smaller and look at 2 peptides in
a box of water (try to get even fewer than 10K waters if you can).
Besides, this is your first REMD right? That's another reason to keep
things simple at the start. If you get good results from this, you can
publish them and them move on to the more complex system if you
believe that you can converge it. Perhaps with your first results in
hand, you could get an allocation from GENCI, Grand Equipement
National de Calcul Intensif, or some similar HPCS organization in
France, to compute the second part.
Chris.
--- original message ---
Dear all,
I come back to you for several questions about the futures
replica-exchange calculations that i would like to perform. The system
of interest will contain 12 peptides (with 7 residues each) and 40000
water molecules, it come from a previous MD performed in NPT ensemble.
With these systems, i would like to study the aggregation process
between the peptides.
After reading several paper about REMD method and playing with the
Temperature generator for REMD-simulations web server
(http://folding.bmc.uu.se/remd/), i suspect that this system is too
big for REMD. Indeed if use the following parameters in the webserver
Pdes 0.2
Temperature range 290 - 600
Number of water molecules 41380
Number of protein atoms 1092
Including all H ~ 1656
Number of hydrogens in protein ~ 240
Number of constraints ~ 1092
Number of vsites ~ 0
Number of degrees of freedom ~ 250464
Energy loss due to constraints 520.59 (kJ/mol K)
I obtain 271 replicas (ouch !!) . If i assume that for each replica
app. 16 CPU, The simulations will be too big and will cost a lot CPU
time.
So my question is can i reduce safely the number of water in system to
reduce the number of replicas ?
For example for 10000 mol of water the number of replicas will be 135.
It is not bad. It is a good option to overcome this limitation.
I have also read the number of replicas can be significantly reduced
by using variants of REMD for example replica exchange with solute
tempering (REST) from Berne and co-workers. Is this method is
implemented in GROMACS ?
Or Can i use the REMD in implicit solvant for example with the coarse
grain OPEP force field as described in Chebaro, et al. (2008).J. Phys.
Chem. B 113(1): 267-274. or by Wang and Voth in J. Phys. Chem. B
112(41): 13079-13090.
Any advices and comments are welcome
Stefane
--
gmx-users mailing list [email protected]
http://lists.gromacs.org/mailman/listinfo/gmx-users
Please search the archive at
http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
Please don't post (un)subscribe requests to the list. Use the
www interface or send it to [email protected].
Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
