Spectroscopic multivariate calibration models require calibration (training) samples that matrix match new prediction sample compositions as well as matching measurement conditions, i.e., matched chemically, physically, and instrumentally. To accomplish this task, calibration sets are often large in sample number to span the expected sources of spectral (matrix) variations. This large range of variability can result in non-linear relationships. One solution to overcome the large number of calibration samples is local modeling. In ideal local modeling, a unique matrix matched subset of calibration samples is selected from a global library for each specific target sample analyte. However, assessing whether or not each sample in a particular calibration set is appropriately matrix matched to new samples relative to the specific analyte is not an easy task. It is well documented that even though calibration samples spectrally match the new sample spectrum, the calibration set is often not fully matrix matched to the new sample composition. For example, the same calibration samples are deemed spectrally similar to anew sample regardless of what the calibration analyte is. This work presents a process that assesses the interactions between a calibration model regression vector and new sample spectra relative to calibration sample interactions. The interactions determine how well each new sample is matrix matched to each calibration sample. The process is general and applicable to other situations such as matching historical batch processing data. Three data sets from the literature are used to demonstrate the functionality of the process. One consists of nuclear magnetic resonance (NMR) spectra for mixtures of three alcohols and the other two contain near infrared (NIR) spectra. One of these data sets is of corn samples with four prediction properties measured on three instruments and the other has mixtures of two alcohols and water measured at multiple temperatures.

Unraveling sample matrix effects for multivariate calibration

2019 Fall National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.