Direct estimation of amylose and amylopectin in single starch granules by machine learning assisted Raman spectroscopy

Abstract

Starch is a fundamental carbohydrate with nutritional and physicochemical properties governed by relative proportions of amylose and amylopectin. Variations in amylose-to-amylopectin ratio significantly influence starch digestibility, texture, glycemic response and dietary fiber functionality. However, conventional techniques such as iodine binding, enzymatic assays and chromatographic separation are often destructive, time-consuming and unable to provide spatially resolved molecular information. Here, we present a non-destructive, label-free approach combining Raman micro-spectroscopy with machine learning to simultaneously classify and quantify amylose and amylopectin within single starch granules. Raman spectra were collected from seven starch varieties and analyzed using multivariate techniques and machine learning including Principal Component Analysis, Linear Discriminant Analysis, Logistic Regression and Support Vector Machines, which enabled accurate discrimination based on spectral features. Key Raman marker bands including 856 and 941 cm⁻¹ for amylose (α-1,4 linkages) and 871 cm⁻¹ for amylopectin (α-1,6 branching) were identified and used in a semi-supervised Multivariate Curve Resolution analysis to resolve overlapping signals and extract pure molecular profiles. Spatial mapping and compositional estimation revealed cultivar-dependent variation, with specific amylose and amylopectin content. This integrated analytical pipeline provides a powerful tool for insitu starch characterization and molecular profiling with potential in food quality assessment, crop selection and industrial starch optimization.