Dynamic UV-Vis monitoring of ternary pharmaceutical photodegradation using multivariate curve resolution and...

Ultraviolet (UV) photodegradation of multicomponent pharmaceutical systems generates complex spectral signals due to overlapping absorption bands and transient intermediates, limiting conventional quantitative approaches. To address this challenge, an integrated chemometric strategy was developed to predict concentration profiles in dynamic ternary mixtures of paracetamol (PAR), naproxen sodium (NPX), and diclofenac sodium (DIC) in spiked river water using Ultraviolet-visible (UV-Vis) spectroscopy. UV-Vis spectra from individual photolysis experiments were analyzed by Multivariate Curve Resolution-Alternating Least Squares (MCR-ALS) to estimate concentration profiles of the parent pharmaceuticals. These profiles were associated with ternary mixture spectra reconstructed by linear combination of the spectral data obtained from the individual photolysis experiments. The combined spectral data, together with experimentally acquired calibration data, served as input to develop a linear Partial Least Squares (PLS) and non-linear Artificial Neural Network (ANN) models, enabling simultaneous prediction of component concentrations. The degree of nonlinearity in the calibration datasets was first assessed using the Non-Linearity Degree (NLD), yielding values of 0.248, 0.398, and 0.214 for PAR, NPX, and DIC, respectively. These results exceeded the 0.05 threshold, indicating significant non-linear behavior in the calibration datasets. Accordingly, the linear PLS model showed limited predictive performance, particularly for DIC. In contrast, the non-linear ANN model achieved higher predictive accuracy, demonstrating its superior suitability for the dynamically evolving multicomponent systems.