pKa disparity between different chemaxon products

I am trying to find out the pKa of a certain molecule. PROBLEM: chemicalize and this website (which is using a chemaxon plug-in) give different pKa values:  

https://epoch.uky.edu/ace/public/pKa.jsp

Which is the more correct? chemicalize gives the 2 pka values of the molecule as ~15 and ~11. the other website gives pka values of the molecule as: ~11 and ~11. 

The molecule in question:

NAME: 1,3-bis(3,4-dichlorophenyl)urea

SMILES: C1=CC(=C(C=C1NC(=O)NC2=CC(=C(C=C2)Cl)Cl)Cl)Cl

PUBCHEM ID: 95783

Hi, 

The reason for the difference between the calculated pKa values is most probably because the two sites use different settings. Now, I don't know what settings the external site uses. I checked the values with a recent version of MarvinSketch, see the attached picture for the values. I consider them the correct ones. 

Daniel

i now have the downloaded marvin bean product. I am confused about whether to consider tautomerisation/resonance in pKa prediction. Because I note that chemicalize doesn't for example. and i note that in the literature the stated predicted pKa values never seem to. so it seems that most are not considering it. which makes me wonder should I or should I not? 

In the case of hyperforin for example. its experimental pKa is 4.8 and marvin predicts its pKa to be 4.67. So very respectably close. BUT - this is if tautomerisation/resonance is NOT considered. if it is the prediction becomes 8.51, which is well off. so tautomerisation/resonance doesn't seem to be helping. 

Please help me understand - in the case of the pKa prediction values differing between using and not using tautomerisation/resonance: which value should I opt for? 

Perhaps it is telling that the chemicalize site doesn't apply tautomerisation/resonance to its pKa prediction? 

Hi, 

Can you send us which version of Marvin you use? Chemicalize doesn't use the tautomerization by default, this is why you get the same results as in Marvin without tautomerization. We are investigating this issue and will get back to you soon. Probably this issue will need a fix. 

Daniel

I downloaded Marvin Beans just a few days ago. so it is likely the most current version. 

to be specific i am using this to deliver pka values:  MarvinSketch 16.3.21

yes - so the paradox for me is that something which presumably is to make the pka prediction better is actually making it worse. seemingly at least (the experimental data could be wrong of course). 

Hi, 

We have investigated this issue and I try to explain what is going on here. The problem why the tautomerization doesn't improve the calculated pK_a is because there is a bug in the tautomerization of hyperforin. I ran a dominant tautomer calculation for the molecule in Marvin and as you can see in the attached picture there are 3 dominant tautomer forms, one double keto form and 2 vinyl forms. Currently we give back the double keto tautomer form as the major form, and when you calculate pK_a with the tautomerization, the input form is converted into this double keto form. This form has a CH acid pK_a value 8.51, this is why you get that as a result.

Unfortunately we don't take into account a special type of H bond for the tautomer calculation, which is missing from our current model. This results in incorrect tautomer distribution and major form. Based on the pK_a values it is probable that in solution there are significant vinyl forms. We will improve our tautomerization model to take this into account. 

Do have any references about the measured pK_a values that we can use maybe?

I hope this helps. 

Daniel

Hi again, 

Do you maybe have any information on which vinyl tautomer form is more stable? Why do you use the form you ran the calculation on?

Thanks, 

Daniel

The experimental pKa value for hyperforin comes from its pubchem entry

https://pubchem.ncbi.nlm.nih.gov/compound/hyperforin

which in turn comes from the following (cut and paste from pubchem, with the reference for it):

pKa = 4.8 (50% aqueous ethanol)

Hi, 

Thank you for your comment. We will provide the bugfix I wrote about ASAP. 

Daniel

Will I have to download a new version of Marvin to get this fix? 

Another point of difference, between experiment and Marvin prediction, is the well known uncoupling molecules FCCP and CCCP.  They both have nitrile groups. Maybe marvin struggles when these are present. 

FCCP (carbonyl cyanide p trifluoromethoxyphenylhydrazone). Its
computationally predicted pKa value is 12.27 while its experimentally derived pKa value is
6.2:

reference: 

THE INTERACTION OF HIGHLY ACTIVE UNCOUPLERS WITH MITOCHONDRIA

HIROSHI TERADA 

Biochimica et Biophysica Acta, 639 (1981) 225-242

Hi, 

This value also needs correction. Thanks for reporting it, we will provide a fix for it soon. 

Daniel

to get fixes - do i need to download a new version from the website? or does the program (present install) receive fixes automatically? 

secondly - have you any idea why the prediction is so far off for FCCP? - could it be that Marvin presently struggles when there is a nitrile group nearby the protobable group? 

Hi, 

No, the program does not update itself automatically. You will have to download the newest version that contains the fix. This will be available next week or in 2 weeks. 

Regarding the question about FCCP, the reason for the off pK_a value is not the nitrile group. 

Daniel

Shifting to logP for a second. What is the name of the algorithm that Marvin uses for logP prediction? so that i can look it up and see what error is associated with it. if you have any references for the latter - that would be good. if not, the name should suffice. 

Hi, 

First of all, the bugfixes for the 2 issues above will be available in the next week's release of Marvin. So I recommend you to update your version next week. 

Regarding your question about our logP calculation,  we use a consensus model based on 3 simple models. You can find the references of the algorithms in our manual :

https://docs.chemaxon.com/display/docs/logP+Plugin

I hope this helps, 

Daniel

RE: logP prediction - can you point me to any literature that gives some sense of how accurate it is by comparing to a test experimental data set. 

Hi, 

For logP we have some in-house validation results here:

https://docs.chemaxon.com/display/docs/Validation+results#Validationresults-logPvalidationresults

In general I recommend you our forum topic for ChemAxon cited articles. In the bottom of this page you can find 2 Nature Review articles for the calculator plugins:

https://www.chemaxon.com/forum/viewtopic.php?p=62803#62803

I hope this helps, 

Daniel

Hi, 

I would like to inform you that we the fixes mentioned above for the hyperforin and FCCP molecules are available in the latest release of Marvin. Please update your version to test them. 

Daniel

I'm now using Marvinsketch 16.6.20 (downloaded today). You say the FCCP (carbonyl cyanide p trifluoromethoxyphenylhydrazone) is fixed - but it is STILL too high. Bear in mind the experimental value is around 6.

And i think the pKa values Marvin returned for 1,3-bis(3,5-dichlorophenyl)urea are wrong also.

I am beginning to think that Marvin (and perhaps pKa programs in general) struggle with NH groups. Especially if close together. 

What do you think of the returned pKa values for 1,3-bis(3,5-dichlorophenyl)urea? are you happy? i think that they are too high. Can you please take a look? thanks mate.  

Hi, 

For the FCCP molecule I get 6.42 for the NH pK_a value as it shows in the picture attached. I took that picture with Marvin 16.6.20. The pK_a value of 6.42 is very close to the experimental value. So I couldn't reproduce your result. Can you check it again?

Regarding 1,3-bis(3,5-dichlorophenyl)urea, yes, the value is probably a bit off, but not very much. We are now looking for an experimental reference and provide correction for the current values. 

Daniel

Daniel - thanks mate: FCCP I have been able to replicate your result now. 

But could you please take a look at the other molecule asap please. i am fairly sure its pKa values are too high.

Hi, 

Ok, so the issue of FCCP is fine, right? We are investigating the other one, stay tuned here. 

Daniel

Yes mate. FCCP is fine now in the new Marvin version. (sorry for my confusion, cleared up now)

Yes - so, the other one. very anxious to hear what you think its pKa values are: because to me they seem too much on the high side. 

Any news on this? - if you still working on it: not to worry. just checking it hasn't been forgotten. 

Hi,

Yes ,  this phenylurea derivative is not forgotten . There is no experimental pK_a value available in the literature for this compound.  

In my opinion this urea derivative  is not ionized under environmental conditions nor even at the human blood pH value.

In neutral form on the other hand it can create strong hydrogen bonds through  the "NH" donor groups of the urea unit. One more interesting structure aspect is the increased lipophilicity generated  by the 4 halogen atoms on the aromatic rings.  These two structure properties may have strong effect on the biological activity too.

The calculated pKa value by Marvin must be considered to be reliable.

Thank you  for reporting these problem(s). 

Cheers,

Jozsi

Its a pleasure. as you can see - i am inputting into marvin and i am using it academically. i have applied multiple times for an academic license and have even queried in the academic license forum. but have never received it or any communication regarding it. could it be possible for you to look into this for me please?  i applied for the academic license with this same account that I write from now.  

Hi!

Hmm! Your comment is important for Chemaxon.

I forwarded it to the relevant Chemaxon's staff.

I think they will clarify this issue soon.

Thanks!

Jozsi

Hi Michael, 

We have asked about your licensing issue the corresponding person, and he is saying that he is working on it. He will get back to you and send the license in the next few days. 

Daniel

thanks mate. 

Hi, 

Have you got your license yet? I hope yes. 

Daniel

Yes mate. i've downloaded it. haven't installed it yet but i assume it will work. will let you know if i have any problems.

Hi, 

Ok, fine. Thank you !

Daniel

Mate. I can't find out how to draw a hydrogen bond in Marvin - to combine 2 molecules (dimerise) via a hydrogen bond. Help! please.    (I will then do a pKa analysis on this dimer)

Hi mate, 

Unfortunately dimerization is not available in Marvin. You can't define H bonds between molecules.

Daniel

This is due to the pK_a values which it is probable that in solution there are significant vinyl forms, so we need to  improve our tautomerization model to make this happen.

Benjamin Hill. Not understood mate. can you elaborate?