Abstract
For large and diverse data sets, simple QSAR methods based on linear and additive models can no longer be applied. In such cases topological methods using descriptors directly derivable from two-dimensional chemical structures provide a useful alternative. The results of such analyses can be used for lead optimization, to guide biological testing and even aid in the design of novel compounds. Various types of topological descriptors and algorithms are briefly discussed. Which of those is to be selected depends on the objective of the investigation and the properties of the data set. Two new methods, LOGANA and LOCON, are discussed in some more detail. With the help of these methods, substructural patterns ("topological pharmacophores") characteristic of compounds possessing a certain biological property can be evaluated. Both methods are designed in such a way that full use can be made of the data handling capacity of computers while maintaining an optimal impact of the experience of the researcher. They are model-free and do not require any mathematical knowledge. While LOGANA deals with semiquantitative or even qualitative biological data, LOCON can be applied to activity data on a continuous scale. The basic procedure in both cases consists in the stepwise combination of substructural descriptors by the logical operations "and," "or" and "not." With a simple example the utility of the methods is demonstrated.