Sustainable analysis of COVID-19 Co-packaged paxlovid: exploring advanced sampling techniques and multivariate processing tools.

The drawbacks of random sampling not only hinder the development of more reliable and efficient methods but also weaken their accuracy, predictive abilities, and validity across several domains. During the current study, a pioneering statistical technique namely, Latin Hypercube Sampling (LHS) was integrated with different multivariate chemometric models namely; Partial Least Squares (PLS), Genetic Algorithm‑Partial Least Squares (GA-PLS), Artificial Neural Networks (ANN), and Multivariate Curve Resolution‑Alternating Least Squares (MCR-ALS). This integration aimed to achieve full data coverage and thereby enhance the predictive powers of these models. Being of clinical significance, Paxlovid(®), a newly co-packaged antiCOVID-19 drug containing ritonavir (RNV)-boosted nirmatrelvir (NMV), was utilized as a study subject to demonstrate the powerful potentials of LHS in enhancing models' robustness and predictive accuracy. The LHS technique was able to provide well-interpreted and informative samples by capturing essential variabilities across the input space without any increase in sample numbers. It was compared and outperformed the random sampling Monte Carlo technique. A comprehensive comparison between the developed models was held where the RMSEP was relatively reduced by 14.1%, 8.9%, 53.1%, and 34.6% for RNV and NMV, respectively using the ANN and MCR-ALS models. Various preprocessing techniques were employed to improve signal quality for PLS construction, yielding superior results (RMSEC of 0.19 for both RNV and NMV) compared to the original, unprocessed spectral data (RMSEC of 0.21 for both RNV and NMV). The Principal Component Analysis score plot was constructed, confirming the consistency of the dataset and the absence of systematic errors, enhancing confidence in the models' robustness. A new hybrid variable selection strategy (GA-ICOMP-PLS) was developed to enhance the robustness and parsimony of the GA-PLS model. Prediction error values of 0.15 and 0.14 were successfully achieved for RNV and NMV, respectively, indicating strong predictive power and generalization. Consistent with sustainability and eco-friendly goals, the current study pioneers the usage of green-blue-white alternatives to conventional analytical methods. A comprehensive assessment was conducted using the "Sample Preparation Metric of Sustainability", the "Analytical Greenness metric for Sample Preparation" and the "Analytical Greenness metric" alongside two solvent sustainability evaluation tools. These evaluations yielded promising results, with green quadrant classification and high scores of 5.89, 0.67, and 0.82 for each metric, respectively, as well as satisfactory t- and F-test values. Moreover, the models achieved outstanding results on the RGB12 metric and Blueness Applicability Grade Index, scoring 96.8% and 82.5, respectively, highlighting their broad applicability, high efficiency, and alignment with eco-friendly analytical practices.