Direct estimation of amylose and amylopectin in single starch granules by machine learning assisted Raman spectroscopy
The study uses Raman micro-spectroscopy in conjunction with machine learning to simultaneously classify and quantify amylose and amylopectin in single starch granules in a non-destructive, label-free manner. It tackles the drawbacks of traditional techniques like enzymatic tests and iodine binding, which are damaging and lack spatial resolution. Raman spectral markers unique to amylose (856 and 941 cm⁻¹) and amylopectin (871 cm⁻¹) were found in this study. PCA, LDA, LR, SVM, and other multivariate and supervised learning algorithms were used in the study to accurately classify different types of starch based on spectral data.
A semi-supervised Multivariate Curve Resolution (MCR) technique was used to fix amylose spectra and separate different chemical components from overlapping Raman data. With amylopectin abundant in the center and amylose near the periphery, it allowed for the spatial mapping of amylose and amylopectin within starch granules, exposing cultivar-dependent distribution patterns. Strong correlation was found when quantitative calculations from Raman-MCR were compared to conventional iodine-binding tests. The study shows a robust analytical process that can perform high-resolution in situ starch profiling without sample damage or chemical labeling. It represents a major breakthrough in polysaccharide compositional analysis and has potential uses in industrial starch modification, agricultural development, and food quality evaluation.
