Qspr-modeling of the Flory-Huggins parameter by quadratic correlation weighting of local graph invariants
Abstract
Results of using descriptors calculated with the so-called correlation weights (CWs) of local graph invariants for modeling of Flory-Huggins parameter are reported. By means of Monte Carlo optimization procedure, values of the CWs which produce as large values as possible correlation coefficient between Flory-Huggins parameter and values of the descriptors on training set have been found. The model of Flory-Huggins parameter obtained by this approach for structures of training set is a reasonable model for the prediction of the parameter values for binary polymer-solvent mixtures from validation set. Statistical characteristics of the model are the following: n = 52, r=0.988, s=0.121, F = 2021 (Training Set); n = 53, r = 0.985, s = 0.123, F = 1666 (Validation Set).