W0040.html

Comparison of Multiple Structural Models Using a Genetic Algorithm. Thomas R. Schneider, Univ. of Gottingen, Dept. of Structural Chemistry, Tammannstr. 4, 37077 Gottingen, Germany, trs@shelx.uni-ac.gwdg.de.

Superposition of a set of related structural models is usually done in an iterative fashion whereby poorly matching regions are excluded until a common core has been found. Especially when more than two models are compared, this approach can be rather tedious and lead to results that are heavily influenced by assumptions made for the initial superposition. In addition, the different levels of coordinate uncertainty in different models or in different regions of the same model are not taken into account systematically.

After normalization to their errors, difference distance matrices[1] allow to objectively access similarities and differences between structural models without any explicit superposition. Recently, a genetic algorithm that rapidly interprets large sets of such matrices in order to identify the part of the molecule that is conformationally invariant with respect to a possibly large set of conformers has been implemented in a computer program [ESCET,2] and its functionality has been demonstrated [3]. Current work concerns the statistical analysis of the difference distance distributions found in the part of the molecule identified as conformationally invariant. Such analysis can be used to validate both the choice of the conformationally invariant region and the accuracy of coordinate uncertainties used in the first place.