Additional Funding Sources
Work was supported by the National Science Foundation under Grant No. CHE-1506417 (co-funded by CDS and E Programs) and is gratefully acknowledged by the authors.
Abstract
Multivariate calibration applied to spectroscopic data is firmly rooted in the field of analytical chemistry. Over the past several decades, numerous methods have been developed to deduce a calibration model to predict new analyte values with sufficient accuracy and precision. These calibration models produce good results when calibration (primary) and new prediction (secondary) samples are measured under similar conditions. However, inherent sample matrix effects and measurement conditions for the secondary samples are often dissimilar to calibration samples resulting in inaccurate and imprecise predictions. To combat this issue, calibration maintenance by model updating can be used to manipulate the calibration model to adapt to the secondary conditions.
Currently, evaluations of traditional and new calibration maintenance methods by researchers are performed without any consideration for the degree of difference between the primary and secondary data sets. Needed is a method that assesses the degree of difference between primary and secondary data sets for a robust evaluation of any model updating method. In order to solve this problem, multiple similarity measures are utilized in this presentation for a fusion consensus assessment of the degree of difference between the primary and secondary spectra assuming equal distributions of analyte values. Results will be shown for spectral data sets of varying similarity.
Fusion of Similarity Measures to Characterize Differences in Sample Matrix Effects
Multivariate calibration applied to spectroscopic data is firmly rooted in the field of analytical chemistry. Over the past several decades, numerous methods have been developed to deduce a calibration model to predict new analyte values with sufficient accuracy and precision. These calibration models produce good results when calibration (primary) and new prediction (secondary) samples are measured under similar conditions. However, inherent sample matrix effects and measurement conditions for the secondary samples are often dissimilar to calibration samples resulting in inaccurate and imprecise predictions. To combat this issue, calibration maintenance by model updating can be used to manipulate the calibration model to adapt to the secondary conditions.
Currently, evaluations of traditional and new calibration maintenance methods by researchers are performed without any consideration for the degree of difference between the primary and secondary data sets. Needed is a method that assesses the degree of difference between primary and secondary data sets for a robust evaluation of any model updating method. In order to solve this problem, multiple similarity measures are utilized in this presentation for a fusion consensus assessment of the degree of difference between the primary and secondary spectra assuming equal distributions of analyte values. Results will be shown for spectral data sets of varying similarity.
Comments
W19