USING REGRESSION MODELS TO ANALYZE QUANTITATIVE STRUCTURE-ACTIVITY RELATIONSHIPS OF PROTOBERBERINE DERIVATIVES

摘要

Abstract. The multivariate linear regression (MLR), the principal component analysis (PCA), the partial least square method (PLS) and the artificial neural network model (ANN) were used to analyze the quantitative structure-activity relationship (QSAR) of protober-berine derivatives. In order to chose the suitable model, its prediction performance on antibacterial activity was surveyed. The result has showed that the neural network and the multivariate linear regression possessed high correlation coefficient R, well fitted to the observed biological activity values and gave high predictive coefficient Q~2. In fact, the linear regression model can be easily used and efficiently made the prediction of the new compounds. Five new compounds having high antibacterial activity were predicted.