The Successive Projections Algorithm for interval selection in trilinear partial least-squares with residual...

Area of Science:

• Analytical Chemistry
• Chemometrics

Background:

• Multi-way data analysis is crucial in modern analytical chemistry.
• Partial Least Squares (PLS) methods are widely used but can be sensitive to irrelevant variables.
• Second-order advantage offers enhanced selectivity in complex mixtures.

Purpose of the Study:

• To introduce and evaluate a novel algorithm, interval Successive Projection Algorithm-N-way Partial Least Squares (iSPA-N-PLS), for variable interval selection in three-way data modeling.
• To assess the performance of iSPA-N-PLS compared to existing methods like N-way PLS (N-PLS) and a Genetic Algorithm (GA) coupled with N-PLS/Residual Bilinearization (RBL).
• To demonstrate the utility of iSPA-N-PLS in handling complex samples with unexpected constituents.

Main Methods:

• Development of the iSPA-N-PLS algorithm, integrating noise reduction and variable selection.
• Modification of the Successive Projection Algorithm (SPA) for interval selection in trilinear PLS.
• Application and validation of iSPA-N-PLS on simulated and experimental datasets (fluorophores, HPLC-UV data).
• Comparison with N-PLS and GA-NPLS/RBL methods.

Main Results:

• iSPA-N-PLS effectively selects informative variable intervals for three-way data.
• The algorithm demonstrated improved quantitative accuracy, evidenced by lower Root Mean Square Error of Prediction (RMSEP) in most cases.
• iSPA-N-PLS showed robustness in the presence of unexpected constituents in test samples.
• Performance was superior to global models and comparable or better than GA-NPLS/RBL.

Conclusions:

• iSPA-N-PLS is a powerful and promising tool for variable selection in second-order calibration.
• The algorithm offers enhanced selectivity and accuracy for multi-way data analysis.
• iSPA-N-PLS provides a valuable alternative for complex analytical challenges.