Artificial Intelligence Helps Scientists Combat Rancidity in Food Products
Have you ever bitten into a nut or a piece of chocolate, expecting a smooth, rich taste, only to encounter an unexpected and unpleasant chalky or sour flavor? That taste is rancidity in action, and it affects pretty much every product in your pantry. Now, with the help of artificial intelligence (AI), scientists can tackle this issue more precisely and efficiently.
We are a group of chemists who specialize in studying ways to extend the shelf life of food products, including those that go rancid. Recently, we published a study highlighting the advantages of utilizing AI tools to keep oil and fat samples fresh for longer periods. Given that oils and fats are commonly found in many food types, such as chips, chocolate, and nuts, the findings of our study could have broad applications in various industries, including cosmetics and pharmaceuticals.
Rancidity and Antioxidants
Food goes rancid when it is exposed to the air for an extended period – a process known as oxidation. In fact, many common ingredients, particularly lipids (fats and oils), react with oxygen. This process can be accelerated by heat or UV light.
Oxidation leads to the formation of smaller molecules, such as ketones, aldehydes, and fatty acids, which give rancid foods their characteristic rank, strong, and metallic scent. Regular consumption of rancid foods can pose health risks.
Fortunately, both nature and the food industry offer a fantastic defense against rancidity – antioxidants.
Antioxidants encompass a wide range of natural and synthetic molecules that can protect food from oxidation. Vitamin C is one example of a natural antioxidant commonly found in foods.
While antioxidants work in various ways, their overall function is to neutralize the processes that lead to rancidity, preserving the flavors and nutritional value of the food for longer periods. In many cases, customers are unaware that they are consuming foods with added antioxidants, as these are typically added in small amounts during the preparation process.
However, it’s not as simple as sprinkling some vitamin C on your food to preserve it. Researchers need to carefully select specific antioxidants and calculate the precise amounts to achieve the desired effect. Combining antioxidants does not always enhance their effectiveness and can sometimes even reduce their protective effect. Finding the right combinations for different types of food involves time-consuming experiments and specialized personnel, driving up costs.
The Role of AI
You’ve probably heard about AI tools like ChatGPT and their capabilities. These systems can analyze large sets of data, identify patterns, and generate useful outputs for users. As chemists, we wanted to teach an AI tool how to search for new combinations of antioxidants.
We selected an AI model capable of working with textual representations, specifically written codes describing the chemical structure of each antioxidant. Initially, we fed the AI a list of approximately one million chemical reactions and taught it basic chemistry concepts, such as identifying important features of molecules. After recognizing general chemical patterns, we further trained the AI using a database of nearly 1,100 antioxidant mixtures described in research literature.
At this point, the AI could predict the outcome of combining any set of two or three antioxidants in less than a second. Its predictions aligned with the effects described in the literature around 90% of the time.
However, when we conducted oxidation experiments in the lab using real lard, the AI’s predictions did not always match the results. This highlighted the challenges of translating computer predictions to real-world laboratory settings.
Refining and Enhancing
Fortunately, AI models are not rigid tools with fixed pathways. They are dynamic learners, allowing researchers to continually feed them new data to improve their predictive capabilities. We discovered that by adding approximately 200 examples from our lab experiments, the AI learned enough chemistry to accurately predict the outcomes, with only slight differences between the predicted and the real values.
Models like ours have the potential to assist scientists in developing better ways to preserve food by determining the most effective antioxidant combinations for specific food types. It’s like having a highly intelligent assistant. Now, we are exploring more effective methods of training the AI model and seeking ways to further enhance its predictive capabilities.
This article was originally published on The Conversation, an independent nonprofit news site dedicated to sharing ideas from academic experts. It is republished here with permission.
This project received funding from the South Carolina Department of Agriculture ACRE Competitive Grant Program. Lucas de Brito Ayres is not affiliated with any organizations that would benefit from this article.
