Identifying Chemical, Physical, and Instrumental Matrix Matched Samples by Leveraging Spectral Model Regression...

Area of Science:

• Analytical Chemistry
• Chemometrics
• Spectroscopy

Background:

• Spectroscopic calibration models require calibration samples that closely mimic new sample compositions and measurement conditions (matrix matching).
• Current methods often rely on spectral similarity, which may not guarantee true matrix matching, leading to poor model performance.
• Assessing the appropriateness of calibration sets for new samples is challenging, especially when considering all chemical, physical, and instrumental factors.

Purpose of the Study:

• To present a novel process for assessing underlying sample matrix interactions between calibration model regression vectors and new sample spectra.
• To enable the identification of calibration samples that are fully matrix matched to new samples.
• To demonstrate the general applicability of the process for various spectroscopic techniques and data types, including unlabeled historical data.

Main Methods:

• Developed a process to analyze the relationship between calibration model regression vectors and new sample spectra.
• Applied the process to two distinct datasets: nuclear magnetic resonance (NMR) spectra of alcohol mixtures and near-infrared (NIR) corn spectra.
• Evaluated the process's functionality using spectral data from multiple instruments and multiple analytes.

Main Results:

• The developed process successfully identified fully matrix-matched calibration samples.
• Calibration samples identified as matrix matched demonstrated the lowest prediction errors for new samples.
• General trends were observed across different data situations, highlighting the process's robustness.

Conclusions:

• The presented process offers a reliable method for identifying truly matrix-matched calibration samples, moving beyond simple spectral similarity.
• Accurate matrix matching is crucial for improving the predictive performance of spectroscopic calibration models.
• This approach has broad applicability in chemometrics, aiding in the development of more robust and accurate analytical methods.