Artificial neural networks aid the design of non-carcinogenic azo dyes

Abstract

This research involves the integration of fuzzy entropies (used in the context of measuring uncertainty and information) with computational neural networks. An algorithm for the creation and manipulation of fuzzy entropies, extracted by a neural network from a data set, is designed and implemented. The neural network is used to find patterns in terms of structural features and properties that correspond to a desired level of activity in various azo dyes. Each molecule is described by a set of structural features, a set of physical properties and the strength of some activity under consideration. After developing an appropriate set of input parameters, the neural network is trained with selected molecules, then a search is carried out for compounds that exhibit the desired level of activity. High level molecular orbital and density functional techniques are employed to establish databases of various molecular properties required by the neural network approach.