reflection radius in somd-freenrg

19 views
Skip to first unread message

Michael Schauperl

unread,
Jul 27, 2020, 10:49:41 AM7/27/20
to Sire Users
Hi,

I was wondering if somd-freenrg is also supporting a similar option to ligandswap's reflection radius? If not, is there a workaround to get a similar behavior? 
If not, what are other options to increase the speed of calculations, expect reducing the number of cycles/moves? 

Thanks for the help,
Michael

Christopher

unread,
Jul 30, 2020, 6:51:25 AM7/30/20
to Sire Users
Hi Michael,

I'm afraid that somd-freenrg doesn't support the reflection radius. This boundary condition only works for Monte Carlo as it is challenging to implement for molecular dynamics (you would need to add support for elastic collisions against an immovable boundary). You could look into coding this support by adding new integrators to OpenMM. This would be a research project, and I guess would take a few months to implement and fully debug.

Speeding up MD is difficult. If the system is stable, you could try increasing the time step, although I think that somd-freenrg is already set to use as large a timestep as is sensible. I've CC'd in Julien who may have more ideas.

  Best wishes,

  Christopher

julie...@googlemail.com

unread,
Aug 7, 2020, 5:45:04 PM8/7/20
to Sire Users
Hi Michael,

There are a few tricks you can use to speed up calculations.

- decrease the size of the solvation box
- truncate the protein before solvation
- use a larger integration time step together with hydrogen mass repartitioning. We have tested this with all bonds constrained in absolute binding free energy calculation setups
(with the following options in your sim.cfg file
 timestep = 4 * femtosecond
constraint = allbonds
hydrogen mass repartitioning factor = 4.0)
It may also work for relative free energy calculations where we normally do not use constraint = allbond, but we haven't used that setup extensively to guarantee it gives correct results.
- You can also get a small speed up for some setups by decreasing the number of cycles and increasing the number of moves, to reduce the overheads associated with cycle initialisation. This depends on the size of your system
- buy a more expensive GPU :-)

Best wishes,

Julien
Reply all
Reply to author
Forward
0 new messages