Lutein, violaxanthin, and zeaxanthin spectrophotometric quantification: A machine learning approach

A machine learning workflow was used to develop spectrophotometric equations quantifying chlorophyll a , b , lutein, violaxanthin, and zeaxanthin simultaneously. Microalgae samples were extracted in methanol following a classical procedure, and no chromatographic separation was applied. To do so, numerous samples with various pigments concentrations (obtained by HLPC) were gathered with their associated visible spectra. The data collected were used to calibrate a machine learning model based on partial least square regression. The best quantification (trade-off between accuracy and over-fitting) was obtained with a seven-feature model (one absorbance and six absorbance derivatives). From a practical perspective, the proposed model is not only calibrated but also validated. Therefore, the equations can readily be used for quantifying lutein, violaxanthin, and zeaxanthin (if high enough). They would significantly shorten the delay in obtaining samples’ carotenoids concentrations compared to liquid chromatography while retaining adequate accuracy (below 10%). Furthermore, the workflow is presented step-wisely so that other scholars may adapt it to their needs (e.g., producing a simpler model focusing only on one pigment). Finally, the data and source files are available in an online repository. Graphical Abstract