MicroNIR spectroscopy and multivariate calibration in the proximal composition determination of human milk

Human milk (HM) is vital for newborns and its importance allied to growing donations has contributed to the expansion of HM banks. However, compositional analysis in HM banks faces many challenges due to traditional methodologies, making it unfeasible for routine inspection. Therefore, this research aims to develop predictive models for the direct determination of the proximal composition in HM samples. The models were developed using spectra acquired with a portable near-infrared spectrometer (MicroNIR) coupled with partial least squares regression and included samples in different lactation phases (colostrum, transition, and mature) and forms (raw and pasteurized). A total of 408 samples were analyzed to give reliability to the models. The performance of the models was estimated by a complete multivariate analytical validation, which indicated satisfactory results with accuracy (represented by the adjust with correlation coefficients ranging from 0.64 to 0.90, and close results for calibration/prediction errors for each parameter). Using the proposed method, moisture, ash, protein, lipids, carbohydrates, and energetic value were successfully predicted. The method is simple, fast, and robust and can be used routinely in compositional analysis of HM banks as an alternative to the traditional methodologies, offering an immediate response with a single and a quick measurement.

Automated summary

This research aimed to develop predictive models for the direct determination of proximal composition in human milk samples. By utilizing a portable near-infrared spectrometer and partial least squares regression, the method successfully predicted moisture, ash, protein, lipids, carbohydrates, and energetic value.