Advances in Atypical FT-IR Milk Screening: Combining Untargeted Spectra Screening and Cluster Algorithms.

Lukas Spieß,Harrie van den Bijgaart,Peter de Peinder

doi:10.3390/foods10051111

Abstract

Fourier-transform mid-infrared spectrometry is an attractive technology for screening adulterated liquid milk products. So far, studies on how infrared spectroscopy can be used to screen spectra for atypical milk composition have either used targeted methods to test for specific adulterants, or have used untargeted screening methods that do not reveal in what way the spectra are atypical. In this study, we evaluate the potential of combining untargeted screening methods with cluster algorithms to indicate in what way a spectrum is atypical and, if possible, why. We found that a combination of untargeted screening methods and cluster algorithms can reveal meaningful and generalizable categories of atypical milk spectra. We demonstrate that spectral information (e.g., the compositional milk profile) and meta-data associated with their acquisition (e.g., at what date and which instrument) can be used to understand in what way the milk is atypical and how it can be used to form hypotheses about the underlying causes. Thereby, it was indicated that atypical milk screening can serve as a valuable complementary quality assurance tool in routine FTIR milk analysis.

Highlights

Fourier-transform mid-infrared spectrometry (FT-IR) is a recognized and widely used method to rapidly determine the compositional quality of raw milk and other liquid milk products
Due to the characteristic effect of the adulterant on the milk’s FT-IR spectrum, mathematical models can be trained to distinguish spectra belonging to adulterated milk from those belonging to normal milk
We show how information in the spectra and meta-data associated with their acquisition can be used to understand in what way the spectrum is atypical and how it can be used to form hypotheses about the underlying causes

Summary

Introduction

Fourier-transform mid-infrared spectrometry (FT-IR) is a recognized and widely used method to rapidly determine the compositional quality of raw milk and other liquid milk products. Targeted methods rely on mathematical models trained to detect the presence—or estimate the quantity—of specific adulterants in the milk. Development of these models requires the adulteration of a sufficiently large and representative collection of milk samples with an adulterant, possibly at different concentrations. Due to the characteristic effect of the adulterant on the milk’s FT-IR spectrum, mathematical models can be trained to distinguish spectra belonging to adulterated milk from those belonging to normal milk. This way, mathematical models have been developed to identify milk adulterated with melamine [7], urea [3], water, starch, sodium citrate, formaldehyde, sucrose, and other adulterants [8,9]. Adulteration with substances (or complex blends thereof) that the mathematical models were not explicitly trained to detect can go undetected

Objectives

Methods

Results

Conclusion