Hello,
I am in the process of implementing some EM/crosslinking modelling using the PMI approach outlined in the tutorial for imp2.10.
I have managed to get my system up and running with PMI2, but I have a couple of very basic questions in regards to the ReplicaExchange0 class in PMI and the monte carlo sampler, as I do not really understand how ReplicaExchange is implemented in PMI.
Would it be possible for you to clarify how simulated annealing and monte carlo are implemented in the ReplicaExchange run in IMP, or point me towards the correct documentation? Does the macro perform a number of simulated annealing and monte carlo steps at each ReplicaExchange frame? Or does this mean that if simulated annealing is enabled, it performs simulated annealing instead of simple monte carlo?
In general, what is the interplay between number_of_frames, monte_carlo_steps and simulated_annealing_minimum/maximum_temperature_nframes?
At the moment, I am running a single process (so no exchange is done, I think).
Additionally, I assume the weight of the restraints is constant throughout the run?
Finally, I plan to implement multistate fitting. In this case, I would have one set of crosslink restraints but the system present in multiple conformations, with an EM density assigned to each state. I was wondering how XL restraints are handled in this case? Is there an OR statement applied?
I hope these questions make sense, as you see I am a bit confused..
Thank you very much in advance,
Andrea
On 2/19/19 4:03 AM, Andrea Graziadei wrote: > In general, what is the interplay between number_of_frames, > monte_carlo_steps and > simulated_annealing_minimum/maximum_temperature_nframes?
number_of_frames is the number of output trajectory frames. Between each frame, monte_carlo_steps of Monte Carlo are carried out. The simulated annealing protocol is very simple (and I wouldn't use it - use REX instead), just setting the Monte Carlo temperature alternately to the minimum and maximum value for the given number of frames.
> At the moment, I am running a single process (so no exchange is done, I > think).
Correct - the number of replicas is the number of processes.
> Additionally, I assume the weight of the restraints is constant > throughout the run?
Sure, but the nuisances of any Bayesian restraints may be optimized.
> Finally, I plan to implement multistate fitting. In this case, I would > have one set of crosslink restraints but the system present in multiple > conformations, with an EM density assigned to each state. I was > wondering how XL restraints are handled in this case? Is there an OR > statement applied?
Each state is (more or less) an independent copy of the system, so in principle yes, each crosslink could be satisfied by either state.
Ben
participants (2)
-
Andrea Graziadei -
Ben Webb