ITMO: Technology for checking dairy products for antibiotics

2023: Introduction of technology to determine antibiotic content and concentration in milk

ITMO scientists have developed technology that automatically detects the content and exact concentration of antibiotics in milk. It is based on electrochemical analysis (a highly sensitive method of detecting the necessary substances in solutions) and algorithmachine learning. The development can protect consumers of dairy products from harmful drugs. It can also be used to analyze other media - for example, to detect unwanted impurities in oil, check the quality of coffee and the authenticity of wine. The university announced this on November 29, 2023.

Traces of antibiotics, which are inevitably used on any farm to prevent and treat diseases in animals, in animal products are a serious problem for manufacturers and consumers. Drug residues can negatively affect human health: for example, cause allergies and antibiotic resistance. Their content in food is strictly controlled by both the state and retailers. Farmers and suppliers must carefully double-check the products before they end up on store shelves.

Typically, test strips are used to analyze milk. However, their accuracy rarely exceeds 70%, and the technique itself allows you to determine only the presence of antibiotics, but not their number. Scientists of the Scientific and Educational Center (REC) of Information Chemistry ITMO have proposed a more reliable technology - it is based on electrochemical analysis and machine learning. The development consists of three components: an electrode sensor, a potentiostat, and a machine learning-based program.

First, we placed the sample on an electrode sensor made of copper, nickel and carbon - upon contact with milk, these substances are oxidized. Then, with the help of a potentiostat (a device for supplying an electric current of a certain voltage), the voltage was "applied" to the electrode. Potentiostat also helped to measure the response from the sensor with the sample: when the metal is oxidized, its conductivity decreases, and the device fixes this in the form of current voltage drops. The less antibiotics in milk, the more the metal on the electrode is oxidized: lactic acid, the product of the vital activity of bacteria, is responsible for this process. Accordingly, the current voltage in the entire system also falls, - said Vadim Belyaev, one of the developers, a graduate student of the REC Infochemistry ITMO.

The researchers conducted a series of laboratory experiments with the five most common antibiotics: streptomycin, penicillin, tetracycline, cefazolin and ceftiofur (the latter is used only in veterinary medicine). Milk powder solutions with individual preparations and their combinations interacted differently with the electrode surface, and their reaction results were recorded by jumps (sharp deviations) on the voltage-ampere characteristics. Scientists have collected all this data into a single dataset for training the machine learning algorithm. The obtained algorithm analyzes the input signals from the electrode and potentiostat and automatically determines the content of antibiotics in the tested samples.

One of the industrial partners of the project is the Galatika group of companies, a large producer of dairy products. According to scientists, their development can be adapted to different tasks. For example, they used a similar technology in creating a technology for detecting unwanted impurities in oil in collaboration with PJSC Gazpromneft. Another idea of ​ ​ the authors is to optimize the methodology for determining antioxidants in wine, as well as the variety and place of growth of grapes from which it is made.

So far, for new applications, the system needs to be significantly rebuilt: change the composition of metals in the electrode and reassemble the library of samples for the algorithm. But in the near future, scientists plan to automate the process of creating new data sets using machine learning models.

We try to generate data using various software packages, such as COMSOL Multiphysics. But my big dream is to prepare a variation autoencoder or transformer by the end of graduate school, which could work only on the basis of these synthetic data. So that we do not have to manually measure each time, for example, zinc in water or a corrosion inhibitor in oil. That is, we want to make the technology easy to use, it itself is trained and adjusted to new tasks, "concluded Timur Aliyev, first author of the study, graduate student of the REC Infochemistry.

The study was supported by the Russian Science Foundation (Project No. 23-16-00224).