主催: 日本化学会情報化学部会
共催: 日本薬学会, 日本農芸化学会, 日本分析化学会, 日本化学プログラム交換機構, 教育システム情報学会(協賛)
p. JP10
QSAR modeling is useful for us to predict some property of unknown chemical compound. But the problems are often in requiring nonlinear modeling. Since artificial neural network is very capable of dealing with nonlinear problems, it can be significant to apply the neural network to QSAR studies. We have developed a software tool for QSAR(Quantitative Structure Activity Relationships) modeling by artificial neural network, called NNQSAR(Neural Network QSAR). NNQSAR can perform learning and predicting the structure-activity data on the same user interface. And NNQSAR has features as follows: (1) two types of modeling, pattern classification and multiple regression (MR) modeling, can be treated. (2) A learning curve can be displayed in a synchronous manner. It makes us easy to visually check the state of convergence and divergence of the learning. (3) The animation of changing synapse weights in the learning process of the network is also available. Thus the changing of the weights can be checked by graphics. (4) In the MR type modeling, the learning and the prediction(evaluation) of a data set of multiplex signals can be performed. It means that the system can treat two or more activities in the learning process (or prediction process) simultaneously. The detail of the system will be discussed with a practical example.