Hierarchy neural networks as applied to pharmaceutical problems

Optimization and prediction are the main purposes in pharmaceutical application of the artificial neural networks (ANNs). To this end, hierarchy-type networks with the backpropagation learning method are most frequently used. This article reviews the basic operating characteristics of such networks. ANNs have outstanding abilities in both classification and fitting. The operation is basically carried out in a nonlinear manner. The nonlinearity brings forth merits as well as a small number of demerits. The reasons for the demerits are analyzed and their remedies are indicated. The mathematical relationships of ANN’s operation and the ALS method as well as the multiregression analysis are reviewed. ANN can be regarded as a function that transforms an input vector to another (output) one. We examined the analytical formula for the partial derivative of this function with respect to the elements of the input vector. This is a powerful means to know the relationship between the input and the output. The reconstruction-learning method determines the minimum number of necessary neurons of the network and is useful to find the necessary descriptors or to trace the flow of information from the input to the output. Finally, the descriptor-mapping method was reviewed to find the nonlinear relationships between the output intensity and descriptors.