Ahnlichkeitsanalyse biologisch aktiver Molekiile mit durch Autokorrelationsvektoren trainierten selbstorganisierenden Karten

Abstract

Topological autocorrelation vectors can be used to estimate similarities of molecular structures. In the following paper we examinedifferent data sets of increasing size and complexity with this measure of similarity. All data sets contain substances with known biologicalactivity on the dopaminergic and benzodiazepine receptors. These two different classes of biological active substances can be separated by self-organizing maps, a kind of neural network well suited for clustering and visualization of similarity. The method is implemented on a massively parallel SIMD computer (MasPar MP-1) which is able to perform this analysis for databases of several thousand substances.