Comments on: Correct protonation state for docking http://www.simbiosys.com/blog/2008/05/17/correct-protonation-state-for-docking/ Addressing the challenges of computational drug discovery Sat, 04 Feb 2012 18:45:03 +0000 http://wordpress.org/?v=2.0.11 by: ChemSpider Blog » Blog Archive » Protonation States and Their Effect on Docking http://www.simbiosys.com/blog/2008/05/17/correct-protonation-state-for-docking/#comment-1477 Thu, 22 May 2008 06:54:10 +0000 http://www.simbiosys.com/blog/2008/05/17/correct-protonation-state-for-docking/#comment-1477 [...] Zsolt came searching ChemSpider for the amino acid structures and found the complexities in terms of charged/stereo forms. But his postinf regarding the “Correct Protonation State for Docking” was an education in itself. If you are engaged in docking experiments at all this is likely a must read. For the rest of us neophytes it’s education! addthis_url = 'http%3A%2F%2Fwww.chemspider.com%2Fblog%2Fprotonation-states-and-their-effect-on-docking.html'; addthis_title = 'Protonation+States+and+Their+Effect+on+Docking'; addthis_pub = ''; [...] […] Zsolt came searching ChemSpider for the amino acid structures and found the complexities in terms of charged/stereo forms. But his postinf regarding the “Correct Protonation State for Docking” was an education in itself. If you are engaged in docking experiments at all this is likely a must read. For the rest of us neophytes it’s education! addthis_url = ‘http%3A%2F%2Fwww.chemspider.com%2Fblog%2Fprotonation-states-and-their-effect-on-docking.html’; addthis_title = ‘Protonation+States+and+Their+Effect+on+Docking’; addthis_pub = ‘’; […]

]]> by: zsolt http://www.simbiosys.com/blog/2008/05/17/correct-protonation-state-for-docking/#comment-1447 Tue, 20 May 2008 15:30:56 +0000 http://www.simbiosys.com/blog/2008/05/17/correct-protonation-state-for-docking/#comment-1447 Andy and Danni, I have added an update to the post in response to your comments. Bottomline is, we can generate the right poses and the right protonation states in eHiTS already -- among other (incorrect) poses with their protonation states. What we need is an accurate QM-supported scoring and ranking of the solutions -- we are working on this, stay tuned... ZZ Andy and Danni,
I have added an update to the post in response to your comments. Bottomline is, we can generate the right poses and the right protonation states in eHiTS already — among other (incorrect) poses with their protonation states. What we need is an accurate QM-supported scoring and ranking of the solutions — we are working on this, stay tuned…

ZZ

]]> by: Danni Harris http://www.simbiosys.com/blog/2008/05/17/correct-protonation-state-for-docking/#comment-1423 Mon, 19 May 2008 19:30:59 +0000 http://www.simbiosys.com/blog/2008/05/17/correct-protonation-state-for-docking/#comment-1423 I agree with all above and echo Andy's concerns about the problem complexity" In addition to the complexity of deducing the 'correct' complementary protonation states of ligand and polar/acidic amino acid sidechains in the binding site cavity one has in some instances: 1)variable conformations of acidic/polar residues in the substrate/inhibitor binding site (e.g. E297 in CYP2C5 crystal structures : 1dt6, 1n6b, 1nr6) and variable water occupancies proximate to those acid residues, 2) connectivity of the ligand and surrounding polar/acidic binding side residues to bulk solvent water via extended hydrogen bond networks (sometimes low-barrier hydrogen bond networks), 3) variable electrostatic influence of enzymatically active species--> that differ with stage in the enzymatic cycle (e.g. P450s). Bottom line: Even when you can frame it with QM/MM and include environment you have to frame the issue for the correct structure ---(amino acids + H2O's) and this may vary with both ligand pose and ligand identity. So doing what you are doing 'on the fly' ZZ is a big step. The next step 'filtering' based on thermodynamic/energetic criteria will be a significant challenge. That being said---let's keep struggling with the problem! :) I agree with all above and echo Andy’s concerns about the problem complexity”

In addition to the complexity of deducing the ‘correct’ complementary protonation states of ligand and polar/acidic amino acid sidechains in the binding site cavity one has in some instances:

1)variable conformations of acidic/polar residues in the substrate/inhibitor binding site (e.g. E297 in CYP2C5 crystal structures : 1dt6, 1n6b, 1nr6) and variable water occupancies proximate to those acid residues,

2) connectivity of the ligand and surrounding polar/acidic binding side residues to bulk solvent water via extended hydrogen bond networks (sometimes low-barrier hydrogen bond networks),

3) variable electrostatic influence of enzymatically active species–> that differ with stage in the enzymatic cycle (e.g. P450s).

Bottom line: Even when you can frame it with QM/MM and include environment you have to frame the issue for the correct structure —(amino acids + H2O’s) and this may vary with both ligand pose and ligand identity. So doing what you are doing ‘on the fly’ ZZ is a big step. The next step ‘filtering’ based on thermodynamic/energetic criteria will be a significant challenge.

That being said—let’s keep struggling with the problem! :)

]]> by: Andy Good http://www.simbiosys.com/blog/2008/05/17/correct-protonation-state-for-docking/#comment-1422 Mon, 19 May 2008 16:15:07 +0000 http://www.simbiosys.com/blog/2008/05/17/correct-protonation-state-for-docking/#comment-1422 Agree completely. This is further exemplified by extending the disuccsion on serine proteases and the catalytic triad histidine. Not only are multiple neutral states possible, in the case of HCV protease the histidine is capable of protonating to form a salt bridge with an acidic ligand. This speaks further to your point regarding protein protonation states being influenced on the fly by different ligands. Whether one can predict these states on the fly is open to debate. Perhaps it would be possible to probe each ambiguous residue with acidic / neutral / basic groups using QM or the like to determine ahead of time which states are energetically acessible before the on the fly screen process begins. It seems unlikely that a purely empircal approach without such an initial callibration would be very accurarate, though I stand ready to be corrected :-) Agree completely. This is further exemplified by extending the disuccsion on serine proteases and the catalytic triad histidine. Not only are multiple neutral states possible, in the case of HCV protease the histidine is capable of protonating to form a salt bridge with an acidic ligand. This speaks further to your point regarding protein protonation states being influenced on the fly by different ligands.
Whether one can predict these states on the fly is open to debate. Perhaps it would be possible to probe each ambiguous residue with acidic / neutral / basic groups using QM or the like to determine ahead of time which states are energetically acessible before the on the fly screen process begins. It seems unlikely that a purely empircal approach without such an initial callibration would be very accurarate, though I stand ready to be corrected :-)

]]> by: zsolt http://www.simbiosys.com/blog/2008/05/17/correct-protonation-state-for-docking/#comment-1383 Sat, 17 May 2008 16:38:14 +0000 http://www.simbiosys.com/blog/2008/05/17/correct-protonation-state-for-docking/#comment-1383 Egon, yes, you used the right phrase "educated guesses", even QM methods will do just that if applied in isolation to the ligand and receptor. My point is, there are tough cases, like HIV-1 and kinases that tend to switch protonation states (both ligand and receptor) upon binding, so for a given ligand what is correct for binding to one receptor is incorrect for binding to another. That is why preparation methods (no matter how sophisticated) are not good enough. You need to analyze each binding pose AFTER docking and make the decision then. But if the docking engine does not work with ambivalent states then it will never place the ligand in a colliding (proton-to-proton or lone-pair to lone-pair) pose so you never find the correct pose (unless you run multiple docking with all possibilities enumerated - very expensive). As for standard benchmark to test on, use the DUD dataset: http://dud.docking.org Of course, that will still not tell you the right protonation state for each pose, but at least you know which ligands are supposed to bind and which ones (although similar) are not. It would be very good to have a test set of PDB complex structures with correctly assigned protonations for some HIV-1 and kinase series. I do not know about the existence of such (PDB-bind has structures and binding affinities but the protonation states were not treated well). One set I know that has been at least eye-balled is the Astex diverse set: http://www.ccdc.cam.ac.uk/products/life_sciences/gold/validation/astex_diverse/ Although it is not focused on the protonation problem. ZZ. Egon, yes, you used the right phrase “educated guesses”, even QM methods will do just that if applied in isolation to the ligand and receptor. My point is, there are tough cases, like HIV-1 and kinases that tend to switch protonation states (both ligand and receptor) upon binding, so for a given ligand what is correct for binding to one receptor is incorrect for binding to another. That is why preparation methods (no matter how sophisticated) are not good enough. You need to analyze each binding pose AFTER docking and make the decision then. But if the docking engine does not work with ambivalent states then it will never place the ligand in a colliding (proton-to-proton or lone-pair to lone-pair) pose so you never find the correct pose (unless you run multiple docking with all possibilities enumerated - very expensive).

As for standard benchmark to test on, use the DUD dataset: http://dud.docking.org
Of course, that will still not tell you the right protonation state for each pose, but at least you know which ligands are supposed to bind and which ones (although similar) are not.
It would be very good to have a test set of PDB complex structures with correctly assigned protonations for some HIV-1 and kinase series. I do not know about the existence of such (PDB-bind has structures and binding affinities but the protonation states were not treated well).
One set I know that has been at least eye-balled is the Astex diverse set:
http://www.ccdc.cam.ac.uk/products/life_sciences/gold/validation/astex_diverse/
Although it is not focused on the protonation problem.
ZZ.

]]> by: Egon Willighagen http://www.simbiosys.com/blog/2008/05/17/correct-protonation-state-for-docking/#comment-1377 Sat, 17 May 2008 13:15:29 +0000 http://www.simbiosys.com/blog/2008/05/17/correct-protonation-state-for-docking/#comment-1377 One can make educated guesses of what the protonation state is, one can also apply some QM-based approach to calculate pKa's... Clearly, it is important... I don't think the CDK currently has an algorithm to do this... Is there a gold standard; a good training set to set approaches on? As you already said, crystallography is not going to help. Databases with experimental pKa database are often proprietary... Any suggestions? One can make educated guesses of what the protonation state is, one can also apply some QM-based approach to calculate pKa’s… Clearly, it is important… I don’t think the CDK currently has an algorithm to do this…

Is there a gold standard; a good training set to set approaches on? As you already said, crystallography is not going to help. Databases with experimental pKa database are often proprietary… Any suggestions?

]]>