Abstract

A QSAR study is reported to predict the binding affinity of a set of 81 modulators for both of human estrogen receptor α and β (ER_α and ER_β). In this study, the derived QSAR models were built by forward stepwise multilinear regression (MLR) and nonlinear radial basis function neural networks (RBFNN), respectively. The statistical characteristics of the external test set provided by multiple linear model (R²=0.814, F=61.277, RMS=0.5461 for ERα; R²=0.600, F=21.039, RMS=0.6707 for ER_β) indicated satisfactory stability and predictive ability of the model built. The predictive ability for ER_β of RBFNN model is somewhat superior: R²=0.7691, F=32.012, RMS=0.5764, and the similar result was obtained for ER_α of the test set: R²=0.7950, F=54.131 RMS=0.3120. Overall, the appropriate results proved the models to be meaningful and useful to predict and virtual screen of the derivatives with high binding activity.

Keywords: Quantitative Structure-Activity Relationships (QSAR), hER_α, hER_β, Multiple Linear Regression (MLR), Radial Basis Function Neural Networks (RBFNN), Binding activity.