What is a good way to set up a replica exchange calculation?

Setting up replica exchange calculations can be challenging. It can be difficult to choose the temperature scheme and to know how many replicas are required to obtain reasonable success rates for exchanging temperatures. The script predict_remd_temp.py, from the Script Center, is designed to deal with these challenges and to potentially reduce the number of replicas needed.

First, download this script from the Script Center. Then set up the replica exchange calculations for your system using the Replica Exchange panel in Maestro (Applications → Desmond → Replica Exchange) and write out the files. The next step is to run predict_remd_temp.py from the command line using the CMS file produced by Maestro:

$SCHRODINGER/run predict_remd_temp.py -t "300 450" -prob 0.2 remd-from-Maestro.cms

where the arguments to -t give the highest and lowest temperatures to simulate and -prob roughly gives the target probability that exchanges will be accepted. This command produces a list of temperatures, and the desmond command to run the replica exchange simulation.

For most solvated proteins the calculation will require much more than 50 processors. If you have sufficient processors you may want to run the simulation using the temperature schedule provided. If not you can try reducing the system size (for example by removing parts of the system or using less solvent), reducing the target acceptance rate or reducing the temperature range. Sometimes these measures are insufficient or the compromises are too great. In such cases the procedure outlined below may help.

Replica exchange calculations are often used to sample conformations. Sometimes the conformational change of interest, even in large systems like solvated proteins, can be considered to be effectively local. The number of replicas needed is related to the number of degrees of freedom being sampled in the simulation. The predict_remd_temp.py script can also increase the masses of selected atoms in the system to the extent that their motions are no longer sampled (they are effectively frozen), which in turn means that the number of replicas can be reduced. The command for doing this is:

$SCHRODINGER/run predict_remd_temp.py -t "300 450" -prob 0.2 --asl ASL-for-unsampled-atoms -o new-REMD.cms remd-from-Maestro.cms

The new CMS file, in which the atoms matching the specified ASL expression are assigned much larger masses, should be used in the subsequent replica exchange calculation. The number of replicas needed can be dramatically reduced if the region that is sampled is significantly smaller than the system as a whole.

Note that the heavier masses are retained in the CMS files produced by the replica exchange calculation. If follow-up Desmond calculations are to be performed using such a CMS file, the heavier masses will persist and these atoms will continue to be effectively frozen. One workaround is to edit out the ffio blocks from the CMS file for the CTs with the heavy atoms and then run the system builder to recreate the ffio blocks with normal masses. To do this you should create a CSB file, my-add-ffio.csb that looks like the following:

{
read_solute_structure heavy-mass-file.cms
write_maeff_file normal-mass-file.cms
}

and then run the System Builder with the following command:

$SCHRODINGER/utilities/system_builder my_add_ffio.csb

Back to Search Results