AM1/d Parameters for Magnesium in Metalloenzymes.

AM1/d parameters are derived for magnesium, optimized for modeling reactions in metalloenzymes. The parameters are optimized with a Monte Carlo procedure so as to reproduce the geometries and energies of a training set calculated with density functional theory. The training set consists of compounds with magnesium coordinated to the oxygen atom of typical biological ligands. Optimization of AM1 parameters without extension to d functions leaves serious errors. The new AM1/d parameters provide a clear improvement in accuracy compared to the standard semiempirical methods AM1 and MNDO/d and will be particularly useful for modeling reactions in large biological systems at low computational cost.