Prediction of permeability of FD-4 through porous poly (2-hydroxyethyl methacrylate) membrane by multiple linear...

The aim of this study was to predict the permeability through porous poly (2-hydroxyethyl methacrylate) (pHEMA) membranes of fluorescein isothiocyanate-labeled dextran molecular weight 4400 (FD-4) as a model of peptide and protein drug movement. Homogeneous standard membranes were prepared by redox polymerization. Permeability data were predicted by an artificial neural network (ANN) as a function of polymerization factors, and the accuracy was compared with that of conventional multiple linear regression (MLR). Good linearity was observed with each model, with the correlation coefficient of a leave-one-out cross-validation (Rcross) being 0.857 for the MLR model and 0.876 for the ANN model. The mean bias and mean accuracy for the ANN were somewhat smaller than those of the MLR. The ANN method provides an accurate quantitative approximation of the permeability coefficient of FD-4, as judged by conventional MLR, and could be applied to prediction of the non-linear relation between polymerization factors and the permeability of FD-4.