3-DIMENSIONAL PATTERN-RECOGNITION FROM MOLECULAR DISTANCE MINIMIZATION

A procedure optimizing the alignment of two three-dimensional molecule s is presented. It uses atomic numbers, positions and radii, and optio nally computed atomic charges. Neither bonds nor connectivities are re quired. The dissimilarity between two molecules is measured with a mol ecular distance, which is minimized under all rotations and translatio ns with a Newton-like algorithm. This molecular distance is the usual norm of a two-components vector. One of the components is the electron ic molecular distance, and the other is the protonic molecular distanc e. Both the electronic and the protonic component are computed with th e same algorithm, which assumes a homogeneous spheres model. The resul ting minimized norm is shown to be an intrinsic molecular distance. Wh en a family of more than two compounds is involved, the intrinsic mole cular distances matrix of the family is built. Various applications ar e presented, including comparisons of X-ray crystallographic data comp ounds, maximal common 3D-substructure searching, comparisons of geomet ry and charge calculations, and quantitative chirality measurement. Th e optimal alignments ale more easily obtained when large atomic radii are selected. The charge calculation algorithm have only a little infl uence on the results.