|
Enantiomeric excess, the ratio between two enantiomers, is an important process variable in chiral catalysis. To increase the efficiency of such processes, this parameter needs to be monitored as close to real time as possible and ideally without elaborate sample preparation. Vibrational circular dichroism (VCD) provides chiral information in a pretreatment free and nondestructive manner. Since classical VCD suffers from a low time resolution, quantum cascade laser (QCL) based VCD was introduced to enable studying more dynamic processes. This significantly improved time resolution enables the use of EC-QCL VCD for monitoring the change of EE e.g. in a chemical process or a chemical reaction. In such applications, the classical approach of human interpretation of individual VCD spectra is no longer reasonable. Hence, chemometric evaluation of VCD spectral datasets is required. In this work we compare accuracy and stability of common multivariate regression algorithms for predicting EE from EC-QCL VCD spectra. Besides classical partial-least-squares regression, modified multiple linear regressions and models derived from chemical knowledge were investigated. We found that a combination of introducing chemical knowledge via spectral descriptor and a reduction of multicollinearity by a ridge regression model resulted in the most stable prediction. Additionally, least absolute shrinkage and selection operator (Lasso) revealed a potential for sensor design involving dedicated QCL arrays focused on a few relevant wavelengths. In summary, a more comprehensive chemometric perspective on QCL-VCD spectra can yield improvements in predictive performance and the shorter measurement times provided by QCL-VCD aid in acquiring datasets of appropriate size.
|
