Transferring multivariate calibration between two optical sensing platforms: NIR and ATR-FTIR
Nika Iurgenson 1 , Yulia Monakhova 2,3, Dmitry Kirsanov 4, 5
1 Institute of Chemistry, University of Debrecen, Debrecen, Hungary
2 FH Aachen University of Applied Sciences, Department of Chemistry and Biotechnology, Jülich, Germany
3 Saratov State University, Institute of Chemistry, Saratov, Russia
4 Institute of Chemistry, St Petersburg University, St. Petersburg, Russia
5 ITMO University, St. Petersburg, Russia
Abstract
The present study investigates the feasibility of transferring multivariate calibration models across two different optical sensing platforms: Near Infrared Spectrometry (NIR) and Attenuated Total Reflection Infrared Spectrometry (ATR-IR). To exemplify this cross-platform transfer, we analyze spectroscopic datasets acquired during the quantification of nicotine, glycerol, and propylene glycol concentrations in e-cigarette refill fluids. Our findings indicate that implementing the Direct Standardization (DS) method enables the interchangeable use of Partial Least Squares (PLS) regression models developed from NIR data with those derived from ATR-IR measurements, and conversely. Prediction errors, measured by Root Mean Square Error of Prediction (RMSEP), increase modestly when applying these transferred models—typically ranging from 1.0 to 1.5 times greater compared to their respective native calibration accuracies. However, the accuracy of transferred models significantly relies upon the specific selection of transfer sets used, occasionally yielding comparable performance relative to directly trained models generated exclusively through the corresponding instrument's spectral output. These results suggest promising potential for creating universally applicable cross-platform multivariate calibration models tailored specifically towards diverse analytical tasks irrespective of underlying measurement methodologies employed.
Speaker
Dmitry Kirsanov
St Petersburg University
Russia
Discussion
Ask question
