Studies on quantitative structure-activity relationships of benzodiazepines using fuzzy neural networks

In this paper, a new fuzzy neural network based on genetic algorithms is pr oposed for quantitative structure-activity relationship (QSAR) studies of b enzodiazepines. The method based on GA+FL+NN allows supervised learning of fuzzy rules from significant examples and is affected unsusceptibly by the problem of local extremes. The network's knowledge base has a linguistic re presentation. This makes it easy for pharmaceutical chemists to understand and interpret. It is possible to introduce current knowledge acquired by re searchers simply by adding one or more fuzzy rules to the network's knowled ge base. Once the fuzzy knowledge base extracted from examples, it can pred ict the pharmacological activity of compounds at a high precision. The obta ined fuzzy rules can also provide useful guidelines for synthesizing new co mpounds with a high pharmacological activity.