Abstract
A quantitative structure activity relationship (QSAR) model has been produced for predicting rejection of emerging contaminants (pharmaceuticals, endocrine disruptors, pesticides and other organic compounds) by polyamide nanofiltration (NF) membranes. Principal component analysis, partial least square regression and multiple linear regressions were used to find a general QSAR equation that combines interactions between membrane characteristics, filtration operating conditions and compound properties for predicting rejection. Membrane characteristics related to hydrophobicity (contact angle), salt rejection, and surface charge (zeta potential); compound properties describing hydrophobicity (log
K
ow, log
D), polarity (dipole moment), and size (molar volume, molecular length, molecular depth, equivalent width, molecular weight); and operating conditions namely flux, pressure, cross flow velocity, back diffusion mass transfer coefficient, hydrodynamic ratio (
J
o/
k), and recovery were identified as candidate variables for rejection prediction. An experimental database produced by the authors that accounts for 106 rejection cases of emerging contaminants by NF membranes as result of eight experiments with clean and fouled membranes (NF-90, NF-200) was used to produce the QSAR model. Subsequently, using the QSAR model, rejection predictions were made for external experimental databases. Actual rejections were compared against predicted rejections and acceptable
R
2 correlation coefficients were found (0.75 and 0.84) for the best models. Additionally, leave-one-out cross-validation of the models achieved a
Q
2 of 0.72 for internal validation. In conclusion, a unified general QSAR equation was able to predict rejections of emerging contaminants during nanofiltration; moreover the present approach is a basis to continue investigation using multivariate analysis techniques for understanding membrane rejection of organic compounds.