Prediction of the maximum absorption wavelength of azobenzene dyes by QSPR tools

The maximum absorption wavelength (lambda(max)) of a large data set of 191 azobenzene dyes was predicted by quantitative structure-property relationship (QSPR) tools. The lambda(max) was correlated with the 4 molecular descriptors calculated from the structure of the dyes alone. The multiple linear regression method (MLR) and the non-linear radial basis function neural network (RBFNN) method were applied to develop the models. The statistical parameters provided by the MLR model were R-2 = 0.893, R-adj(2) = 0.893, q(LOO)(2) = 0.884, F = 1214.871, RMS = 11.6430 for the training set; and R-2 = 0.849, R-adj(2) = 0.845, q(ext)(2) = 0.846, F = 207.812, RMS = 14.0919 for the external test set. The RBFNN model gave even improved statistical results: R-2 = 0.920, R-adj(2) = 0.919, q(LOO)(2) = 0.898, F = 1664.074, RMS = 9.9215 for the training set, and R-2 = 0.895, R-adj(2) = 0.892, q(ext)(2) = 0.895, F = 314.256, RMS = 11.6427 for the external test set. This theoretical method provides a simple, precise and an alternative method to obtain lambda(max) of azobenzene dyes. (C) 2011 Elsevier B.V. All rights reserved.