| 摘要 |
PURPOSE:To accurately analyze a structure active correlation using the learning function of a neural network by providing a jumping input layer for inputting specified data directly to the intermediate layer of the neural network. CONSTITUTION:A jumping input layer 18 for directly inputting prescribed numeric data to every layer of an intermediate layer 12 of a neural network 16 is provided. In this case, the numerical data at the permeation process, the metabolism process and the combination reaction process of a medicine which already turns out active or inactive are defined as the learning data for a jumping input, and by using activity or inactivity as basic input learning data and an educator signal except it, the whole network including the jumping input layer is made to be learnt by a back propagation method. The weight of the network is analyzed after this learning, and especially the weight of the network from the input layer to the intermediate layer is analyzed. Thus, it is possible to know the degree of the control of the jumping input data to each construction process, and to judge a factor controlling the structure activity accurately. |
