Question

I am sorry there was already a question earlier on this, but I did not get what I wanted, hence I am asking as another question. Please answer.

Do OpenBabel, OpenEyeChem generate 3D co-ordinates using the same Force Fields(MMFF94) or do they use different force fields?

Is there like a best force field ??

Pubchem documentation says it uses OpenEyeChem's OMEGA (MMFF94).

Well if I take a CID and get their 3D coordinates from PubChem and compare those co-ordinates with 3D co-ordinates generated using open babel --gen3D option I find the co-ordinates values differ. Are they not supposed to be same ? Why/why not ?

I actually have to further run Gaussian or GAMESS on these, so would not my results differ? Would that matter at all or no ?

Thanks for ur time.

Answer 1

There are many reasons why the coordinates differ:

• molecules have multiple conforms, a force field does not favor one
• the optimized conformer is defined by the starting structure
• given numerical calculations involved, solutions are not to be expected numerically identical
• direct comparison of coordinates is not possible anyway, because of rotational freedom

Running a force field before more computationally demanding methods (GAMESS, Gaussian) is typically done, to first get a quick-and-dirty solution, as getting to that solution in the more accurate method is much more time consuming. That said, the latter will typically further optimize the geometry, so the exact quick-and-dirty solution does not matter. But, the first will define the conformer, and the latter will typically not escape that solution.

There is not really a best force field; most are optimized towards a certain data set, or class of molecules. Make sure to read to papers published on the force field you use, to find areas where they do not perform nicely. An Open alternative force field, is the UFF force field implemented in Jmol.

Depending on your application, you may want to look at conformation sampling methods too.

Answer 2

Beyond what Andrew and Egon have said, remember that there are vast differences in implementation between Open Babel and Open Eye as far as 3D coordinate generation:

1. How do you create the coordinates in the first place (before running MMFF94)? Do you have a set of ring templates? Do you lay out the atoms like a graph algorithm, or do you use a matrix method with distance constraints, or some combination?
2. How many iterations do you use for the force field? What kind of optimization algorithm do you use?
3. How do you search for a low-energy conformer? Randomly? Monte Carlo? How many conformers do you sample?

As both comments explain, there is almost certainly no "one true conformer" for a large, complex molecule.

Answer 3

As said already, there is not one algorithm creating "the conformation". There are many ways to create conformations. Some possibilities for a small molecule conformer generation are

• exhaustive enumeration. This will still deliver a discrete number of solutions, since typically a small but finite torsion around each rotatable bond will get used. The results are always reproducible, but a lot of time is wasted in creating irrelevant ("bad") structures.
• deterministic rule for creating conformations. The idea is to create diverse conformations and to reach "the best" conformations (whatever that means) in a short time. The results are reproducible (deterministic).
• stochastic sampling for creating conformations. Here the reproducibility depends on the seeding of a random generator. In general it can not be granted that "the optimal" structure will get produced, still, in practice those algorithms perform very well and show a good quality/runtime behavior.

Of course, is this is a oversimplification of all the research already done in this area. For more algorithmic and performance details I would point to the following reviews: Conformational boosting, macrocycles, drug molecues.

The energy scoring and force-field used opens a completely other field of research, which is of particular relevance for defining the correct solvation model and entropy terms. One recent review about solvation is Brooks2008. For understanding differences in force-fields I would point towards the original articles about the MMFF94 and OPLS force-fields.

Finally, if we start talking about (first-principle) conformations of proteins or homology models of proteins things become even more complicated. For good reviews I would point to the CASP challenge.

Answer 4

There is no such thing as "the" conformation for a structure. Even a diatomic structure has thermal energy keeping it from the lowest potential energy conformation, and with larger compounds there may be potential energy barriers making the lowest energy conformation inaccessible.

The different tools use a number of methods in order generate which is hoped to be a good representative structure.

But let's suppose they all used MMFF and suppose they all were searching for the minimum energy confirmation. The problem leads directly into the Levinthal paradox, in that the brute force takes intractable time. Energy space has many local minimas, so the differences in search methods naturally lead to different solutions.