Understanding Minimize Results

[Hao Shen]

Hello,

I am running --minimize on some docked poses. For the most part smina is giving reasonable results. I have some protein ligands that have high RMSDs

I have some runs that look like:
Affinity: -5.92551 -10.48481 (kcal/mol)
RMSD: 6.04010

While other runs (same exact command, just running it again) look like:
Affinity: -7.55955 -10.32162 (kcal/mol)
RMSD: 1.14702

Checking these two results, it seems like smina has flipped the molecule from the original pose in the former case.

I tried doing several things to try to get a more deterministic output such as increasing exhaustiveness and setting a fixed seed, but I always see this stochasticity. Is there a way to make results deterministic (and in this case I would prefer the latter result)? Other things I tried: increase num_modes -> nothing different, still really fast, does it even create more modes? increasing factor -> slowed it down a lot, more deterministic results, but seemed to give intermediate results. From my understanding, smina should find the local minimum from the initial pose, so I don't see why there should be two modes that I am seeing.

Thanks!

[David Koes]

--minimize should be deterministic. Docking is only deterministic if you set the random seed.

[Hao Shen]

I just tried it with two other docked poses which seem to be deterministic, woo! However, the complex I was referring to in the beginning is still not deterministic. The only arguments I'm passing in are -l, -r, --flexdist 4, --flexdist_ligand, --minimize. I have attached the ligand and protein files if anyone would like to take a look yourself. Any ideas as to what might be causing this?

[David Koes]

One of your flexible residues (831) has an alternate location that hasn't been removed. This means the residue contains atoms that are overlapping one another (there are two versions of the residue). I haven't exactly tracked down how this is inducing non-determinism, but my guess is that it results in a divide by zero or infinity somewhere in the calculation which result in undefined behavior. If you remove the alternate residue you get deterministic minimization.

[Hao Shen]

Hello, I have fixed my issue, the 1M17 had a duplicated residue 830 AASP and BASP that was messing everything up. When I only kept one of those residues, smina converged properly. Maybe the energy was blowing up or something?

[Hao Shen]

Woops, didn't see that you commented already, but I'm glad we reached the same conclusion. Thanks a lot!