No, but the README (https://sourceforge.net/projects/smina/files/README) provides a klunky hack for getting something like covalent docking (last paragraph).
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I have a couple of pose solutions in which a link atom is present in both protein and ligand having exactly the same coordinates (In this case a S from a CYS residue). I have rescored the solutions using smina with the flag --score_only. I chose to not use the quasi-covalent mode from smina because all my poses have exactly this overlap. My thought was that this overlap would be a systemic error and would not affect the relative ranking of poses. Scores could be obtained without any errors and none had an affinity prediction below zero which was to be expected. However, ranking the scores from high to low and comparing these with experimental data delivered a high correlation (in terms of spearman). This was done for ~60 proteins with ~20 ligands each. Each ligand having ~5 predicted poses. I also compared these spearman correlations to other scoring functions and smina outperformed them for about halve of the proteins. Do you have a suggestion why higher (more positive) predicted affinities gave a good correlation?
KR,
Yannick van Dijk
Last edit: Yannick van Dijk 2022-10-21
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
There's some correlation between whatever is inducing the higher score and affinity. Having you tried minimizing to the local optimum and using those scores?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
I don't know if I understand you correctly. I am using smina to rescore the poses that I have. I have obtained the poses using GOLD. These poses were thus minimized by GOLD already. We are not using any flexibility for the protein except for hydrogen rotation. Smina is used as an extra scoring function outside the GOLD scoring function package to check whether it could give a better ranking as compared to experimental data.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
You have different scoring functions. They will not have the same local minima. For example, if GOLD prefers hydrogen bonds to be closer than smina's defaults, they will contribute more to the repulsion term and result in higher (more positive) scores, even though hydrogen bonds are favorable. You need to minimize to get a local optima for the score to make sense. I'd also recommend using gnina if your goal is effective rescoring.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thanks for your quick answers! I see what you mean. However, I think this will not work for this case because of the covalent docking mode. If I would refine the pose, the link atom will be dangling somewhere it should not be. This is the reason why we kept the pose as is. Keeping it this way, the linkatom (both present in the protein and ligand) would remain at the same position for every generated pose from GOLD. Our hypothesis was that keeping the pose as is would then result in a systemic error. When we would minimize it again, the error would be intractable. It was surprising that smina outperformed other scoring functions compared with experimental results because the affinity gave an negative correlation as expected. I thought maybe there is something about the terms smina uses which gave it such a correlation so we could trace back why. I also rescored some poses and rescored it similarly for non-covalent inhibitors and this gave reasonable quantities. Do you think it is possible to change the covalent link atom to a dummy atom like Boron (not present in both the ligand and protein) and then apply a very strong guassian potential between them?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Yes, that's how we fake covalent docking currently. Usually the local optimum isn't very far away though - you can check the minimizedRMSD and if it is small assume the compounds is still essentially in covalent range.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Does Smins support covalent docking?
No, but the README (https://sourceforge.net/projects/smina/files/README) provides a klunky hack for getting something like covalent docking (last paragraph).
Thanks David
Dear David Koes,
I have a couple of pose solutions in which a link atom is present in both protein and ligand having exactly the same coordinates (In this case a S from a CYS residue). I have rescored the solutions using smina with the flag --score_only. I chose to not use the quasi-covalent mode from smina because all my poses have exactly this overlap. My thought was that this overlap would be a systemic error and would not affect the relative ranking of poses. Scores could be obtained without any errors and none had an affinity prediction below zero which was to be expected. However, ranking the scores from high to low and comparing these with experimental data delivered a high correlation (in terms of spearman). This was done for ~60 proteins with ~20 ligands each. Each ligand having ~5 predicted poses. I also compared these spearman correlations to other scoring functions and smina outperformed them for about halve of the proteins. Do you have a suggestion why higher (more positive) predicted affinities gave a good correlation?
KR,
Yannick van Dijk
Last edit: Yannick van Dijk 2022-10-21
There's some correlation between whatever is inducing the higher score and affinity. Having you tried minimizing to the local optimum and using those scores?
I don't know if I understand you correctly. I am using smina to rescore the poses that I have. I have obtained the poses using GOLD. These poses were thus minimized by GOLD already. We are not using any flexibility for the protein except for hydrogen rotation. Smina is used as an extra scoring function outside the GOLD scoring function package to check whether it could give a better ranking as compared to experimental data.
You have different scoring functions. They will not have the same local minima. For example, if GOLD prefers hydrogen bonds to be closer than smina's defaults, they will contribute more to the repulsion term and result in higher (more positive) scores, even though hydrogen bonds are favorable. You need to minimize to get a local optima for the score to make sense. I'd also recommend using gnina if your goal is effective rescoring.
Thanks for your quick answers! I see what you mean. However, I think this will not work for this case because of the covalent docking mode. If I would refine the pose, the link atom will be dangling somewhere it should not be. This is the reason why we kept the pose as is. Keeping it this way, the linkatom (both present in the protein and ligand) would remain at the same position for every generated pose from GOLD. Our hypothesis was that keeping the pose as is would then result in a systemic error. When we would minimize it again, the error would be intractable. It was surprising that smina outperformed other scoring functions compared with experimental results because the affinity gave an negative correlation as expected. I thought maybe there is something about the terms smina uses which gave it such a correlation so we could trace back why. I also rescored some poses and rescored it similarly for non-covalent inhibitors and this gave reasonable quantities. Do you think it is possible to change the covalent link atom to a dummy atom like Boron (not present in both the ligand and protein) and then apply a very strong guassian potential between them?
Yes, that's how we fake covalent docking currently. Usually the local optimum isn't very far away though - you can check the minimizedRMSD and if it is small assume the compounds is still essentially in covalent range.