QSPR-modelling of octanol/water partition coefficient by means of the nearest neighboring codes correlation weighing

A descriptor calculated with the so-called correlation weights of local graph invariants has been suggested. As local graph invariants, Morgan vertex degrees of zero, first, and second order, and the so-called Nearest Neighboring Codes, have been used. The numerical values of the correlation weights have been calculated by the optimization procedure, which leads to an as large as possible correlation coefficient between hydrophobicity (logP, octanol/water) and the above descriptor values, on the training set. The predictive ability of such correlation has been tested with structures of the test set, the best model being obtained by means of the correlation weighing of the Nearest Neighboring Codes.