Two fundamental types of information are obtained from any molecular mechanics study, namely the minimum value of the strain energy, and the structure associated with that minimum. Agreement between the energy-minimized and experimental (crystallographic) structures has often been used as the primary check on the validity of the force field and to refine the force field further, but often little predictive use has been made of the structures obtained. As force fields have become more reliable, the potential value of structure predictions increases. More importantly, when no unequivocal determination of a structure is available by experimental methods, then structure prediction may be the only means of obtaining a three-dimensional (3-D) model of the molecule. This is often the case, for instance, in metal-macromolecule adducts, and structures obtained by molecular mechanics can be a genuine aid in the visualization of these interactions. In this chapter we consider the ways in which structure prediction by, or aided by, molecular mechanics calculations is used, and also point to current trends and developments.
7.1 Accuracy of Structure Prediction
The accuracy of molecular mechanics models in terms of reproducing known structures varies enormously, depending on the quality of the force field. The best-established force fields for metal complexes are for hexaaminecobalt(III) cations. Many such complexes have been modeled, and the agreement is generally AE0.01 ร
in terms of bond lengths, AE2 in terms of valence angles, and AE5 in terms of torsion angles [95, 104, 171, 173, 244]. It is unlikely that this level of agreement can be improved on substantially because the disagreement is either close to the level of precision of the crystallographic experiment, or close to the level of variation observed in crystallographic structures as a result of crystal packing and hydrogen bonding. The same is true for some other transition metal complexes with amine and a few other simple ligand systems, but in many other cases the reliability of structural predictions is lower.