Geometry tool

User d57717f4c7

21-07-2013 20:29:50

I know it has been discussed and I read all the relevant topics but can you please advise on how to use the Geometry tool to calculate minimum projection radius of simple molecules like CO, CO2, etc.?


The two attached Geometry results show that the tilting of the green axis results in considerable overestimates of the minimum projection areas. I would expect the minimum projection radius for these two to be in the order of the Van der Waals radius of C (1.7A), which corresponds relatively well with the 1.61 of the reported Length perpendicular to max area (1.61A). Interrestingly, the minimal and maximal projection radii of pure C are -1.


Can you help me out here?

User d57717f4c7

22-07-2013 16:01:41

Also, you can easily check what happens to the minimum projection radius of (straight) polyaromatic rings, like


naphthalene, anthracene, tetracene, hexacene (which should all have the same minimum projection radius with benzene).


 


None of them is calculated correctly (not even benzene), again because of the tilting of the green axis (the minimum projection plane). Geometry calculates a minimum projection radius of 3.420A for benzene, and 5.205A for hexacene (with all kinds of values for the other PAHs).


If I am not doing something wrong, it means that the estimations of the Geometry tool are far from accurate and you might want to correct it ASAP or remove it alltogether until fixed.

ChemAxon 60613ab728

23-07-2013 11:59:38

Dear Sir/Madam,


Thank you for your comment regarding the accuracy of the Geometry Descriptor Calculator tool.


This is a known issue that due to the approximations used during the geometry search and the projection calculations, the accuracy of the results may vary in case of highly symmetric molecules and in some other corner cases.
Please find corresponding posts in the forum, e.g:
https://www.chemaxon.com/forum/viewpost34222.html#34222 


If you insist to a more accurate or even analytical solution of the problem, we can discus further refinement of the tool and the details of the development in email. Please contact us on the disco-support@ and sales _at_ chemaxon.com email addresses.


I can give you a reference, where the accuracy of the present method was sufficient to use the method for drug discovery research:


Louis C. Martineau - Simple thermodynamic model of unassisted proton shuttle uncoupling and prediction of activity from calculated speciation, lipophilicity, and molecular geometry


In the present version, in order to increase accuracy, you can set the optimization limit to very strict. I expect you would get better results for the challenging cases like hexacene.


Best Regards,


Miklos 

User d57717f4c7

23-07-2013 14:02:31

Actually, when I tried to run the Geometry tool on hexacene with the "very strict" setting I got a message saying


" Bug, optmization did not converge"


The error message in detail is here:


 


ERROR: 


Cannot run plugin: Geometry


[BUG] Optimization was not converged.


STACK TRACE:


java.lang.UnsupportedOperationException: [BUG] Optimization was not converged.


at chemaxon.calculations.ProjectedSurfaceArea.optimize(ProjectedSurfaceArea.java:153)


at chemaxon.calculations.ProjectedSurfaceArea.run(ProjectedSurfaceArea.java:237)


at chemaxon.calculations.Geometry.calculateMoleculeProjection(Geometry.java:336)


at chemaxon.marvin.calculations.GeometryPlugin.run(GeometryPlugin.java:703)


at chemaxon.marvin.plugin.CalculatorPluginManager.runPlugin(CalculatorPluginManager.java:200)


at chemaxon.marvin.plugin.CalculatorPluginManager.runPlugin(CalculatorPluginManager.java:184)


at chemaxon.marvin.plugin.CalculatorPluginManager.runPlugin(CalculatorPluginManager.java:107)


at chemaxon.marvin.plugin.CalculatorPluginManager.callback(CalculatorPluginManager.java:77)


at chemaxon.marvin.plugin.PluginManager.callback(PluginManager.java:92)


at chemaxon.marvin.common.swing.MolPanel$6.run(MolPanel.java:3755)



After you have the thermodynamically favorable tructure, finding a cylinder impostor (or the minimum and maximum bounding cylinders) is not a difficult task. I am actually developing a code in Matlab to see if I can fix this using the xyz geometry I get from Marvin (another bug that you might be able to help me with: when Marvin exports xyz files it ommits the H atoms; can we fix that?). I don't have access to the Marvin code, but if it does indeed try to find the minimum bounding cylinder, then that works well but it has an error in the estimation of the data centerline (which determines the direction of the cylinder). In the cases presented here, estimation of the xyz coordinates centerline is quite trivial (it is the mean of the xyz coordinates). So I don't see how you can get this wrong (I guess you don't try to find a cylinder impostor?).   


Interestingly, I cannot replicate my own result for hexacene (I run with strict setting for the optmization, with which it now crashes). With the normal setting in optimization it gives a minimal projection radii of 3.652 (which is still not correct but not as bad as before).


All the molecules I worked with here are in reality 2-dimensional (there is nothing asymmetrical going on in the z direction). This makes it trivial to test the Geometry tool. Other people working with molecules in 3D geometries, have no way of assesing whether the estimates from the Geometry tool are true or not. But knowing that one might get projection radii that are about double the correct values, tells me that one should be extremely cautius when using this tool. I would put it as a first priority in your future revisions. 


Will send an email to see if we can figure this out for my current study.

User d57717f4c7

23-07-2013 14:19:21

One last note. Using the normal setting in the optmization I got the following structure.


 


You can visit chemicalize (http://www.chemicalize.org/structure/#!mol=hexacene&source=fp) and see that the reported minimal projection area is larger... Finding that green centerline in Marvin appears to be a matter of luck :-)

ChemAxon 60613ab728

23-07-2013 16:13:09

Thank you for your valuable comments. As I mentioned, the present solution is based on approximation. Thus, it has many known limitations, just as you pointed out. 


Chemicalize may use different parameters than Marvin. And it is not a surprize to me that it suffers from the same limitation for the same example.


We are looking forward to hearing your further suggestions in more details. Please contact us in email if the improvement of this tool is crucial for you.


Best regards,


Miklos