Three-point multivariate calibration models by correlation constrained MCR-ALS: A feasibility study for quantitative...

Area of Science:

• Analytical Chemistry
• Chemometrics
• Spectroscopy

Background:

• Quantitative analysis of complex mixtures often relies on multivariate calibration methods like Partial Least Squares (PLS).
• PLS requires a large, representative set of calibration samples for accurate predictions, which can be costly and impractical.
• The selection of calibration samples significantly impacts the predictive accuracy of multivariate models.

Purpose of the Study:

• To evaluate Correlation Constrained Multivariate Curve Resolution (CC-MCR) as an alternative regression method for quantitative analysis in complex mixtures.
• To demonstrate the feasibility of using a minimal three-point calibration strategy with CC-MCR.
• To compare the performance of CC-MCR with traditional PLS methods when using limited calibration data.

Main Methods:

• Employed Correlation Constrained Multivariate Curve Resolution (CC-MCR) for iterative regression and optimization under constraints.
• Utilized a three-point calibration strategy, selecting samples with minimum, maximum, and average concentrations.
• Assessed the method's performance through case studies involving mixtures with interfering species.

Main Results:

• CC-MCR models provided reasonable predictions in quantitative analysis even with only three calibration samples.
• Achieved satisfactory predictions with relative errors ranging from 3-15% across various case studies.
• Demonstrated good agreement between CC-MCR predictions and results from classical PLS models built with larger datasets.

Conclusions:

• CC-MCR offers a viable and efficient alternative to traditional methods like PLS for quantitative analysis in complex mixtures.
• A three-point calibration strategy is effective for building reliable CC-MCR models, overcoming the limitation of requiring numerous calibration samples.
• CC-MCR is particularly advantageous when the number of available calibration samples is limited.