Neural Network Model for Predicting Anticancer Activity of Pyridopyrimidines Derivatives
Molecular structures of pyridopyrimidines derivatives as known as dihydrofolate reductase (DHFR) inhibitors were investigated by using the neural network method. Based on the molecular connectivity, molecular connectivity index and molecular electronegativity distance vectors of 32 pyridopyrimidine derivatives were obtained. Among these parameters, three optimized structural parameters 1χ3χ and M17 as the input neurons of the artificial neural network were selected by step-wise regression. Then a 3:4:1 network architecture was employed and a satisfying neural network model for predicting anticancer activity (lg1/C) was constructed with the back-propagation (BP) algorithm. The total correlation coefficient R and the standard deviation S were 0.925 and 0.336 respectively that showed significantly nonlinear relationships between lg1/C and three structural parameters. It was concluded that the predictions of BP neural network are better than those of methods in the literatures.