search

MobleyLab / blues

Applications of nonequilibrium candidate Monte Carlo (NCMC) to ligand binding mode sampling
https://mobleylab-blues.readthedocs.io/en/latest
MIT License
33 stars 16 forks source link

protocol work turned nan after RandomLigandRotationMove #173

Open VincenzoChen opened 2 years ago

VincenzoChen commented 2 years ago

I have tested the t4-toluene example and it worked well. But when I use my own system , it hasn't work quite well. In this system, the ligand is greatly larger than toluene and is in the solvent but not in protein pocket. And I has not freeze any groups. The protocolWork turned nan after performing TranslationMove in NCMC show as following:

#"Iter" "Progress (%)"  "Step"  "alchemicalLambda"  "protocolWork"  "Speed (ns/day)"    "Time Remaining"
ncmc: 0 5.0%    25  0.05    39.08851538888597   0   --
ncmc: 0 10.0%   50  0.1 70.96068947520838   26.7    0:05
ncmc: 0 15.0%   75  0.15    94.11311782093316   28.2    0:05
ncmc: 0 20.0%   100 0.2 108.8464813136671   26.5    0:05
ncmc: 0 25.0%   125 0.25    108.8464813136671   26.5    0:04
ncmc: 0 30.0%   150 0.3 108.9467082762027   26.4    0:04
ncmc: 0 35.0%   175 0.35    108.9467082762027   26.2    0:04
ncmc: 0 40.0%   200 0.4 108.9467082762027   25.8    0:04
ncmc: 0 45.0%   225 0.45    108.8464813136671   25.9    0:03
ncmc: 0 50.0%   250 0.5 109.04693523873831  22.5    0:03
Performing RandomLigandRotationMove...
ncmc: 0 55.0%   275 0.55    nan 16.8    0:04
ncmc: 0 60.0%   300 0.6 nan 13.6    0:05
ncmc: 0 65.0%   325 0.65    nan 11.8    0:05
ncmc: 0 70.0%   350 0.7 nan 10.8    0:04
ncmc: 0 75.0%   375 0.75    nan 10.1    0:04
ncmc: 0 80.0%   400 0.8 nan 9.52    0:03
ncmc: 0 85.0%   425 0.85    nan 9.08    0:02
ncmc: 0 90.0%   450 0.9 nan 8.73    0:01
ncmc: 0 95.0%   475 0.95    nan 8.45    0:01
ERROR: [simulation._stepNCMC] Particle coordinate is nan

So I wander how this phenomenon comes into being. Maybe it is the clash between ligand and solvent that cause this problem?

davidlmobley commented 2 years ago

This is a good question and I'm not sure the answer. You're saying that you're doing this in solvent, with no protein around? Would you be able to share a 2D structure or SMILES string for your ligand?

VincenzoChen commented 2 years ago

There is a protein and ligand in the system, but the ligand is in the solvent. The 2D structure of ligand are shown in picture. Screenshot from 2022-05-07 09-29-58

davidlmobley commented 2 years ago

OK, it looks like you're going to get extremely large energies from this for multiple reasons: (a) those charged groups are going to have very large interaction energies with their environment; (b) this is so large and flexible that almost any rigid-body rotation of it in a binding site will result in steric clashes so severe that BLUES won't be able to recover.

I think this is out of the domain of applicability of the current versions of BLUES.