Convergence with increasing exhaustiveness
Scoring and Minimization with AutoDock Vina
Status: Beta
Brought to you by:
dkoes
Menu
▾
▴
I love Smina, but I found something little bit worrisome: if you increase exhaustiveness of the docking, the docking result can be erratic. This happened with one of my systems one flex sidechain at 100 or 200 exhaustiveness (can't remember now). I don't know if the same is true for rigid receptor. Basically, the program works, but there are some (algorithm?) issues which could pop up in unexpected places. Just to let you know.
Can you be more descriptive about what the issue is? This is a stochastic algorithm, you should expect variability in the results unless you set the random seed.
That was very odd, but the story was repeating for several of my systems, I will describe "the worst" case. I don't have exact numbers, I would have to dig into files or perhaps re-run some runs, but let's say at exhaustiveness 50 docked energy was -2.5, at exhaustiveness 100 E=-1.2, then at exhaustiveness 200 B.E. recovered al ittle, let's say, -2. I looked at the generated pose of the worst "-1.2" case, and strangely the ligand was stuck in an obviously higher energy pose, stacked between some residues where it didn't have much space, and certainly not in the X-ray-like position. (I was using Vinardo, by the way).
Last edit: Visvaldas Kairys 2020-12-05
Are you running each time with the same random seed? If not, you should not expect reproducible outcomes and there is nothing unusually with what you are describing.
Unless this is a small fragment, none of those are good scores.
I tried changing seeds, to no avail, the ligand gets stuck in the high energy. Sorry, I was mistaken, I found the actual data, the good score was -6.5, the bad -1.2.
The question is whether the lower exhaustiveness and higher exhaustiveness runs had the same random seed. If not, then you should not expect the same outcome - the lower exhaustiveness may have just gotten lucky.
Now it seems to work fine. I don't understand why I had so hard time tuning exhaustiveness previously. Maybe it does have to do with the random seeds. When the problem happens again, I will let you know. Sorry for the disturbance...
Sorry this is a rhetorical comment. My student just now was doing docking with the recently updated "mainstream" AutoDock, and at exhaustiveness =100 and vinardo scoring he got variation of about 0.5 kcal/mol for the same ligand/receptor pair with different random seeds. Smina on the other hand gives 0-0.1 kcal/mol variation for the ligands he has tested. So, praise to Smina, and warning regarding the "vanilla" AutoDock (probably needs to be debugged).
There is a bug in vanilla Vina where it "optimizes" the positions of hydrogens despite hydrogens not participating in scoring. It is possible this is responsible for the extra variance. The option
--flex_hydrogensre-enables this bug for exact compatibility with Vina if you want test this hypothesis.