[Openmm-team] Application OpenMM questions

[Christopher Bayly]

Hello OpenMM-team,

I have installed and tested (with testInstallation.py) the binary distributions for Linux and MacOSX and they work. That is great, thank you.

I was running amber sander and pmemd for many years but I am a newb to OpenMM, so here are a few questions:

I ran simulatePdb.py on my Mac with only 
DYLD_LIBRARY_PATH=/opt/openmm/lib
ie I did not specify cuda things because I would like to use OpenCL. How do I know if it did use it or not? Should I add something to the library path to ask it to use OpenCL? My Mac Quadbook with a regular nvidia card finished the 10000 steps in 191 min, so that's 50 steps/min on your villin-waterbox case, but not knowing your code or that example I can't judge if that's GPU fast or x86_64 fast.

I looked at the output (10 snapshots, 1 per 2 ps) and waters look like they're detaching. Are they just diffusing into neighboring periodic images or maybe there isn't a box? The script does specify "nonbondedMethod=PME" but where is the boxsize specified? The CRYST1 line in the pdb header, or does OpenMM infer it from the xyz coords input?

When starting up protein-ligand complexes with modeled ligand poses (and just for startup in general), harmonically restraining nonH coords to their starting positions is a common way to prevent the initial "twang" (due to bad clashes) from screwing up the configuration. I didn't see such harmonic positional constraints in the API... is that available?

Christopher Bayly
OpenEye Scientific Software

PS - please say hi to Sherm from me