Comparative study of partial least squares and multivariate curve resolution for simultaneous spectrophotometric determination of pharmaceuticals in environmental samples

Abstract

The potentials of partial least squares regression (PLSR) and multivariate curve resolution alternating least squares (MCR-ALS) are evaluated for the simultaneous determination of diclofenac (DCF), naproxen (NAP), mefenamic acid (MEF) and carbamazepine (CBZ) as target analytes and gemfibrozil (GEM) as an interference in synthetic and real environmental samples. The analysis of first-order UV-Vis spectra is performed using PLSR with different variable selection methods, which include variable importance in projection (VIP), recursive weighted partial least squares (rPLS), regression coefficient (RV) and uninformative variable elimination (UVE), and using MCR-ALS with correlation constraint (MCR-ALS-CC). The obtained statistical parameters in terms of relative error (RE), regression coefficient (R²) and root mean square error (RMSE) were satisfactory for the calibration and validation sets. Furthermore, in real environmental samples, the obtained statistical parameters of PLSR using VIP and rPLS as well as MCR-ALS-CC were reasonable considering the heavy overlap of the target analytes and complex sample matrices. In general, PLSR showed a better performance for the determination of analytes in samples that are free of interference or contain calibrated interference(s). On the other hand, MCR-ALS-CC allowed the accurate determination of analytes in the presence of unknown interferences and more complex sample matrices.