Clean 3D problems, clean 3d methods
When I do a 3D clean in Marvin 5.1.5 at your Web site, the two CH3 groups end up at about a 45 degree dihedral angle, not the appropriate 90 degree dihedral angle.
|<?xml version="1.0" ?> |
atomID="a1 a2 a3 a4 a5 a6 a7 a8 a9 a10 a11 a12 a13"
elementType="C C C C C H H H H H H H H"
x3="-0.3496770489448852 -0.33187453949201595 -0.3715981507134086 -1.3849077944388577 -1.5976911649079888 0.5139928992591869 0.32347934075526374 -1.5938219568650076 -2.3258288180680435 -1.0260252586698098 -1.4841277491723133 -2.4136789540536565 -1.8745623055316893"
y3="0.550594965135553 0.580824330743454 0.6289151140483628 1.399510253946418 0.48212344745609964 0.5502696594919676 0.14526133718422835 0.9523782868013115 1.4390132039322376 2.41949373342929 0.9588610875441963 0.9881271610995186 -0.5633239389865583"
z3="1.812307982415561 0.5467765814910885 -0.7177034625924388 -1.4034037791655614 2.5393366238549646 2.3048083815689275 -1.238691786261633 -2.3771126579927864 -0.8426049967547455 -1.5559003108374818 3.5147881486074564 2.0106473957987667 2.6903343017839147"
<bond atomRefs2="a1 a2" order="2" />
<bond atomRefs2="a2 a3" order="2" />
<bond atomRefs2="a3 a4" order="1" />
<bond atomRefs2="a1 a5" order="1" />
<bond atomRefs2="a1 a6" order="1" />
<bond atomRefs2="a3 a7" order="1" />
<bond atomRefs2="a4 a8" order="1" />
<bond atomRefs2="a4 a9" order="1" />
<bond atomRefs2="a4 a10" order="1" />
<bond atomRefs2="a5 a11" order="1" />
<bond atomRefs2="a5 a12" order="1" />
<bond atomRefs2="a5 a13" order="1" />
Thanks for the bug report. The Dreiding force field does not specify energy component for this dihedral. In the future we are planning to add this to out implementation. When it is ready we will notify You in this topic.
being very interested in this structure field, I have evaluated the five "Clean 3 D" methods given in MarvinSketch (5.2), see attached image. (All views are along the linear C=C=C bonds).
Obviously the methods give different results. The "Fast Build" method produces the best result, nearly 90 degrees.
Question: What is the purpose to have these five methods ?
Question: What is the purpose to have these five methods ?
The five methods covers a couple of use-case scenarios:
- Fine build generates coordinates for the given structure. Implicit hydrogenes are respected during coordinate generation, however they are not returned, thus the given topology is kept intact
- Fine with hydrogenize returns the eplicit hydrogenes too.
- Fast build uses a differenct coordinate generation approach which in some cases can be faster than the fine build.
- Build or optimize invokes a coordinate generation for 2D/0D structures or a geometry optimization for already 3D structures
- Gradient optimize invokes a geometry optimization starting from the given coordinates.
thank you for your answer.
To evaluate these five methods I used the example L-Isoleucine coming from the template library. After applying the Clean 3D methods I used "Tool - Geometry - lowest energy conformer Never " to get an analysis on the quality of the Clean 3D result.
As can be seen in the attached image, the results are very different. Only method (2) "Clean 3D Fine with hydrogenize" gives an acceptable result, which is nearly the same as the "real" structure optimizations "Geometry lowest conformer Always" or the calculations of "all" conformers, these have the most stable lowest energy value of 60 kJ/mol. The configuration is correct as 2S,3S as can be seen in the right part of the image.
The result of method (5) is not 3D, it is still a planar structure as the structure on the canvas.
Is something wrong in this evaluation? If not, I suggest to delete the methods (1), (3), (4) and especially (5).
Modern computers are so fast, that the calculation time should not be a criterion for methods giving bad results.
Best regards, Hans-Ulrich
Thank you for the prudent analysis of this functionality. I can not examine right now the discrepancies between the energy found by Geometry plugin and the two Fine builds, however i would like to adress two of Your notices:
|The result of method (5) is not 3D, it is still a planar structure as the structure on the canvas. |
Is something wrong in this evaluation?
Gradient optimize is the force field based geometry optimization. As we discussed it before this finds the nearest local energy minimum from the starting geometry. This method should not (however for some structures it does) "break" from 2D: there should be no force in a 2D structure pointing out from the plane! So this behavior is normal.
|If not, I suggest to delete the methods (1), (3), (4) and especially (5). |
For me it seems that - supposed the discrepancies between generated energies fixed - method "Fast build" needs further attention. Due to the several possible use cases the other methods are needed to keep.
"Fast build" in the current form is obnoxious, however the often debated need for a "chemically not necessarily valid but fast 3D coordinate generation" can be justified. However our very limited available capacity for work with Clean3D now (and in the foreseedn near future) undoubtedly need to be focused on the improvement of chemically valid functionality.
It is definitely true, however one or two orders of magnitude difference in speed is devastating (between a fast 3D generation and a multiconformer "global minimum" approach) when you consider the fast 3D screening of multi-million molecule sized databases or even bigger virtual synthesis libraries!
|Modern computers are so fast, that the calculation time should not be a criterion for methods giving bad results. |
To summarize the above, thanks again for the substantial comparison, which presumably will be used as a validation protokoll when this issue will be addressed. I aggree that the "fast build" option in the current form not necessarily acceptabble, however i heavily disaggree with the suggestion about remove them.
thank you for your detailed answer. I cannot agree all points, but before continuing this discussion, I will evaluate some more examples and then I will answer to your statements.
The next example is Biphenyl. This molecule seems to be evident and simple, but it is not. There is a lot of literature on the structure, both experiment and calculations (f.e. F. Grein, J. Phys. Chem. A 2002, 106, 3823-3827). In solid state it seems to be planar due to packing forces. In solvents and in a melt there is a torsion angle of the two rings of about 33 to 38 degrees. And this torsion leads to chiral molecules, particularly for 2,2'-subtituted biphenyls, and leads to the possibilty to use these a catalysts for asymetric syntheses.
Please see the image what MarvinSketch i doing with this molecule.
a) First the super result produced by the Tool - Conformers - all conformers: As expected by experiment the molecule is neither planar and nor it has a torsion angle of 90 degrees. It is "twisted" with an angle of 60 degrees. Having this angle not exactly at the experimental value is not so bad as it seems. To repeat it, the important result is, the molecule is chiral and has 2 enantiomers ! This is a really important result for a software being available in the internet-applet for all user evaluating chemistry. And the conformers tool produces 2 enantiomers with of course equal energy.
b) Method (1) gives the same angle as (2) but the result of (2) is obviously much better than (1).
c) Method (3) produces the wrong planar structure with the worst energy value.
d) Method (4) produces the torsion angle, but the energy is bad.
e) Method (5) is, as mentioned in the earlier posts, not a "Clean 3D" method and leaves the molecule in the given planar structure. Interestingly, the energy value is the same as for (3) "Fast Build".
This molecule is an interesting test for calculation methods. Some of the different force field models available produce also the wrong planar structure as energy potential minimum, and at least one semiempirical method, too. "Ab Initio" quantum chemical calculations differ in the magnitude of the torsion angle and have a hot discussion on it.
The next example will be posted soon.
now one Biphenyl example more. This molecule belongs to the 76 "parameter molecules" of the paper describing the parametrization procedure of the "Dreiding force field". (Of course you know the literature reference: J. Phys. Chem. 1990, 94, 8897). It is also a Biphenyl derivative. The X-Ray structure shows a torsion angle of 37.5 degrees, very close to the value for the unsubstituted Biphenyl. In this case the molecule becomes also non-chiral, when the torsion angle is zero or 90 degrees. So it is important to get a reasonable torsion angle.
The results of the 5 methods are corresponding to the results for the unsubstituted Biphenyl. Method (3) "Fast build" produces once more an unacceptable 3D structure.
The next molecule will be a very important one.
the next molecule is a very important one: D-Glucose. I have learned about 50 percent of the molecules in our biosphere have this fragment (Cellulose). Here MarvinSketch gives very good results:
Drawing the correct wedge-formula is easy and then to produce reasonable 3D coordinates is really very productive, only a few mouse-clicks and you have it. Super !
Here method (3) is not so bad as before.
Method (2) is once more much better than the other methods, and using the MarvinSketch GUI you cannot measure a calculation time difference.