So far I have got a first set of charges and parameters for bonds and angles. The dihedral angles are obviously trickier. One can model the exocyclic bonds (methoxy and phenyl) through potential energy scans, this is laborious but otherwise straightforward, but what to do about torsions in the tropolone and B-ring? I am unsure how to generate appropiate ring conformations. I would appreciate some hints here.
I always perceive the dihedral fitting as the least laborious part of a parameterization project. For the water interactions, one must submit several interaction calculations, with a requirement to think about the right water placement for each of them. For the hard degrees of freedom, putting together a molvib input is significant work. But for the scans, both Gaussian and Q-Chem can do a whole Potential Energy scan with just one input. And the fitting is usually relatively straightforward as well (though this might be one of the exceptions).
But to answer your question: tropone has a very low normal mode (53 cm-1 at MP2/6-31G(d)) that roughly corresponds to an envelope distortion, ie. the most apparent motion when visualizing this mode is the carbonyl carbon (and oxygen) moving out of the plane of the ring. This mode is important because its very low frequency makes the ring very flexible; indeed, in the colchicine model compound O-methyltropolone, the ring minimizes to a planar or a nonplanar minimum depending on the initial C-C-O-C torsion. These kind of relatively large conformational changes associated with low modes tend to be relatively poorly reproduced when just fitting molvib, so it would be advisable to perform a concerted scan along this mode on tropone before moving on to methyltropolone and scanning the C-C-O-C torsion. I was going to send you an existing example of such a concerted scan, but the examples I found were not optimally relevant for the current problem at hand, so I instead made the attached tropone input.
It should be noted that the 2 dihedrals I'm using to drive the scan are not the only ones that change significantly during the scan; it's geometrically impossible to change less than 4 dihedrals in a ring without big changes in bond lengths. But since this is a relaxed scan, I just let the other dihedrals adapt. I have reasons to believe that the resulting concerted motion in this particular case will closely mimic the actual low mode (and that's more than sufficient for the purpose of dihedral fitting). Conversely, I don't think we could get away with driving the ring scan by only one dihedral because the resulting motion would be qualitatively different from the one we're interested in.
Finally, purely based on considerations of the degree of freedom, this one scan would not be sufficient to fully capture the flexing behavior of this ring. However, given our choice of atom types, I think we will get away with it, especially if we also consider the molvib and/or the conformations that are visited during the scan.
I always perceive the dihedral fitting as the least laborious part of a parameterization project. For the water interactions, one must submit several interaction calculations, with a requirement to think about the right water placement for each of them. For the hard degrees of freedom, putting together a molvib input is significant work.
The beauty of natural products (as opposed to traditional "drug-like molecules") is that they inhabit a completely different region of chemical space. "Drugs" are generally flexible, but natural projects are rigid, they are scaffolds from which functional groups protrude.
Consequently, if the hard degrees of freedom are a little off, it doesn't matter at all, provided the overall shape is retained. But this flexing behaviour of the tropolone ring in colchicine, it seems as if that is the key to binding, and that one must be treated correctly, as must the charges.
Colchicine doesn't look particularly rigid to me - certainly not more rigid than your average drug. I expect the flexing of the central 7-membered ring (is that what you call the B-ring?) to be quite important too - though potentially somewhat easier to capture with the existing parameters.
Saturated n-membered rings have n - 3 torsional degrees of freedom/low modes. For each in-ring double bond, one can subtract one, so that would formally give 7 - 3 - 2 = 2 df/lm. It is possible that the steric interaction and π conjugation between the A and C ring restrains things further, but that would still leave the tip of the ring (if one considers the bond between A and C as the base) free to "envelope flip" in the same fashion as tropone. This motion would have an important effect on the position of the amide substituent.