Molecular sensors for traceability and quality control
Towards "functional foods": molecular sensors
With the SANUM project we investigated the use of molecular sensors for traceability and quality control.
Nutrition-related health problems such as food allergies, obesity, diabetes and cardiovascular disease have grown in epidemic proportions. Furthermore, they are also causing a heavy toll on our society and our health systems. For this reason, the European Commission has launched a challenge to ICT companies, the Horizon Prize, for the development of a cheap and non-invasive mobile solution that would allow users to measure and analyze food directly (https://ec.europa.eu/info/research-and-innovation/funding/funding-opportunities/prizes/horizon-prizes/food-scanner_en). Up for grabs: € 1 Million! Three companies have shared this award.
We wondered why not try to use these new technologies in the Nurideas R&D? And then integrate these molecular sensors into the traceability and quality control we were developing for micro agribusiness companies?
In 2017 we presented a research and development project in response to the call “Filiera Biomed” of Sardinia Research (POR FESR 2014-2020). Indeed, we wanted to try to answer this question.
The SANUM project objective was the study of new functions related to food traceability and personalized nutrition. With the intention of integrating them into the ecosystem of products and services we develop at Nurideas. And thus increase the level of innovation of our company pipeline.
The project included a scientific validation process for the sensors divided into four phases:
- to begin with, collection of an adequate number of food samples from dairy farms;
- following that, analysis of the samples at the NMR and GC-MS spectrometry laboratories of the University of Cagliari;
- at the same time, analysis of the same samples with molecular sensors currently available on the market;
- finally, validation of the results obtained.
We have tested 2 of the 3 sensors that were winners in 2016 of the European Horizon Price Food scanner award. We had tow objectives. Firstly, to analyze the sensitivity level of the sensors and validate how they function. And then understand if they could be used by micro and small agri-food businesses.
Both sensors use NIR (Near-Infrared Spectroscopy) technology but operate in different wavelength ranges (SCiO: 750-1040 nm; Tellspec: 900-1700 nm).
From a technical point of view, the sensors consist of a light source that irradiates the food sample. Then an optical sensor (spectrometer) converges the rays of light reflected by the sample. After that it returns a spectrum that is used for the final analysis of the sample.
In the SANUM project we involved three dairy farms and a multifunctional farm located in different parts of the territory. Between December 2017 and July 2018, we analyzed various types of milk (goat, sheep and mixed sheep-goat) and cheese.
We analyzed 67 samples corresponding to 16 different products and we obtained a total of 2319 spectra. To put it in another way, these spectra can be considered as a fingerprint of each product.
On the collected samples we have run data-mining analysis. To do that we used the tools offered by the “TheLab” application on the Consumer Physics website (developer of the SCiO sensor).
TheLab allows you to perform a PCA analysis (Principal Component Analysis) on the spectra and view the results in graphical form.
The figure shows an example of a PCA made on a group of 4 different cheeses, that we can visually identify without problem.
In PCA analyses we look for “clusters” or clouds of data that have the same properties. We made this kind of analysis on all the samples. It has allowed us to filter all the anomalous spectra (outliers). Therefore we could proceed and clean up the data for the subsequent creation of the models.
A data-mining model is created by applying an algorithm to the data. It is a set of data, statistics and schema that are then applied to new data. As a result, the model generates predictions and inferences about relationships. With the samples and data collected we created a classification models to test the validity of the approach. We used the model generation tool in “TheLab” using the same 4 cheeses shown in the PCA. In the figure we see the expected performance of the classification model (the so-called “confusion matrix”).
However, to develop a more precise model, we will need to analyze a larger number of samples for each type of cheese.