Experimental study of non-linear second-order analytical data with focus on the second-order advantage

Area of Science:

• Analytical Chemistry
• Chemometrics
• Computational Chemistry

Background:

• Complex samples often contain unexpected interferents that challenge traditional quantitative analysis.
• Second-order instrumental data offers a 'second-order advantage' for analyte determination, but requires sophisticated algorithms.
• Existing methods like unfolded partial least-squares/residual bilinearization may struggle with non-linear data.

Purpose of the Study:

• To evaluate a novel algorithm combining artificial neural networks (ANNs) with post-training residual bilinearization (PT-RB) for analyte quantitation.
• To demonstrate the efficacy of the ANNs-PT-RB approach across three diverse experimental systems with complex matrices.
• To compare the performance of ANNs-PT-RB against traditional methods for improved prediction accuracy and precision.

Main Methods:

• Application of ANNs-PT-RB to non-linear second-order instrumental data from three distinct analytical systems.
• System 1: Absorbance-pH matrix measurements for pharmaceutical determination.
• System 2: Kinetic-based determination of iron(II) using catalytic effects.
• System 3: Fluorescence excitation-emission matrix analysis for antibiotic quantitation.

Main Results:

• The ANNs-PT-RB method achieved high prediction accuracy and precision in all three studied systems.
• Successful recovery of analyte profiles was observed, even in the presence of unexpected sample components.
• The ANNs-PT-RB approach significantly outperformed the unfolded partial least-squares/residual bilinearization model.

Conclusions:

• The combination of ANNs with PT-RB is a powerful and versatile tool for quantitative analysis of analytes in complex samples.
• This method effectively leverages the 'second-order advantage' inherent in non-linear instrumental data.
• The ANNs-PT-RB approach offers superior performance compared to linear methods for non-linear analytical data.