Machine Learning-Enabled Quantification of Fucosylated Human Milk Oligosaccharides in Human Breast Milk by Benchtop ¹ H NMR Spectroscopy

Human milk oligosaccharides (HMOs) are bioactive components of human breast milk (HBM), but their concentrations vary with maternal secretor phenotype and lactation stage, making individual HMO quantification analytically demanding. Here, we present a reference-guided benchtop 60 MHz NMR workflow integrating chemometrics and machine learning to quantify major fucosylated HMOs in HBM. A total of 111 HBM samples from 37 donors across lactation stages were analyzed. Using 800 MHz NMR-derived concentrations as references, predictive models were developed from benchtop NMR spectra for 2'-fucosyllactose (2'-FL), 3-fucosyllactose (3-FL), and lacto-N-fucopentaose I (LNFP-I). Unlike conventional peak-fitting-based NMR quantification, this workflow recovered HMO-specific quantitative information from highly overlapped 60 MHz carbohydrate signals that were not directly resolvable in authentic HBM. Elastic Net yielded practical models, particularly for lower-abundance 3-FL and LNFP-I. These results support benchtop NMR combined with machine learning as an accessible platform for HMO screening.