rvkrishnan30
commented
1 year ago Hi @Yoshanuikabundi , as suggested during the follow-up workshop, I loosened the convergence criteria for ForceBalance to check if I get any speed improvement without loss of accuracy.
By changing both convergence objective and convergence step from 0.01 to 1.00, the speed improved from ~40 hrs to ~19 hrs without much loss in accuracy as from the K values and the resultant torsion plots.
I am sure even this can be further improved in terms of speed without issues and suspect something is taking much longer than it has to slowing down the entire FB fitting process. I see this kind of a behaviour with a few other unrelated ligand molecules too.
Could you please suggest if there is a better way to debug and narrow down to what's causing the issue, and also possibly what other parameters can be tuned to avoid this behaviour.
Below is my example input file passed to ForceBalance.
$options ffdir forcefield penalty_type L1 jobtype optimize forcefield force-field.offxml
maxstep 10 convergence_step 1.00 convergence_objective 1.00 convergence_gradient 0.01 criteria 2 eig_lowerbound 0.01 finite_difference_h 0.01 penalty_additive 1.0
trust0 -0.25 mintrust 0.05 error_tolerance 1.0 adaptive_factor 0.2 adaptive_damping 1.0 normalize_weights False constrain_charge false
priors ProperTorsions/Proper/k : 6.0 /priors
$end
$target name ligand_fragment_around_22_24 weight 1.0
type TorsionProfile_SMIRNOFF mol2 ligand_fragment_around_22_24.sdf pdb ligand_fragment_around_22_24.pdb coords scan.xyz writelevel 2 attenuate 1
energy_denom 1.0 energy_upper 10.0 $end
j-wags
mark-mackey-cresset
Description
During the BespokeFit followup workshop, @rvkrishnan30 pointed out that the S1A ligand takes >24 hours during ForceBalance. I haven't been able to reproduce this yet because my XTB calculations are slow for some reason but I will soon.
Reproduction
I'm attempting to reproduce in this gist
Output
@rvkrishnan30 any details you could provide on this would be great :)
Software versions
conda list?