Look into as an alternative for constrained minimization
Might need to experiment with force constants– work out right force constant to get the desired behaviour
Eventually look into working into FB if it works really well
Things to look out for – RMSD to the QM structures not too high, the torsion angle itself not veering 5 degrees from the grid point.
ForceBalance currently does the following:
constrains dihedral
Sets 1 kcal/mol/AA restraint on the rest of the atoms
We could look into
restraining dihedral
Possibly setting a higher restraint on the rest of atoms if RMSD is too high in minimization?
Next steps
Benchmarking current constraint protocol (0 mass thing)
RMSD of the molecule at each grid point to QM
might be script that does this already in torsion repo
Investigating the restraint protocol
RMSD of the molecule at each grid point to QM, with different force constants to see which gives best behaviour (high enough that torsions don’t deviate more than 5 degrees from grid points, and RMSD doesn’t deivate “too much” from original QM)
FC low enough (hopefully?) that super high forces can relax a little
Might need two, one for dihedral, one for the same kind of restraint as ForceBalance uses for other positions
Have a look this week
ForceB
benchmarking infrastructure repo?
progression going
Next week will try to set up bigger runners
smee
XFF re-fit. 250k records. 350 steps in 12 days. Using 140+ GB RAM
Looking into JHorton’s batching solution
Benchmark at 350 steps?
datasets
lipids
ball is in Julianne’s court
sulfur
Fixed bugs, have server running
chemical perception work – re-fitting force fields