[Rdkit-discuss] Strange MMFF94 Optimization Results?

[Patrick A. <ps...@bu...>]

Hey there RDKitters,

I have been generating conformers in RDKit, optimizing them using MMFF94,
and sorting them by their energies. For some tests, I have been using
caffeine. But I seem to have some strange results, and I wonder if anyone
knows why.

I have attached two images of an MMFF94 optimized caffeine conformer. In
them, the oxygen in the top right corner is the primary strange thing I see
(although it doesn't seem to be very planar either, which may be somewhat
strange).

Note that the oxygen is sticking out of the plane of the molecule. This is
the lowest energy conformer generated and optimized by RDKit (and other
conformers low in energy are similar to it - with the oxygen sticking out).

If I take that exact molecule and MMFF94 optimize it in Avogadro, all the
atoms move to be in the same plane (including the oxygen that was sticking
out). So the problem goes away if I MMFF94 optimize it with Avogadro.

So the question is: why are RDKit and Avogadro giving different results for
the MMFF94 optimization? And why does RDKit's MMFF94 push the oxygen out of
the plane so much?

In RDKit, I am using the C++ function
RDKit::MMFF::MMFFOptimizeMoleculeConfs().

My parameters are the mol (with all the conformers in it), an empty result
vector, 1 thread, 1000 maximum optimization iterations, "MMFF94" for the
mmffVariant, 100.0 for the nonbonded threshold, and true for
ignoreInterfragInteractions.

Let me know if anyone knows why.

Thanks,
Patrick