“There are many advantages of using AI in mycotoxin research”
How can AI help mycotoxin research and management? This is the title of the presentation given by Ine van der Fels-Klerx, principal scientist at Wageningen Food and Safety Research, during the 15th World Mycotoxin Forum. In this interview she will further explain about the role Artificial Intelligence plays in mycotoxin research and management.
MYCOTOXINS 2025: Utilising technology to detect & mitigate – read all articles
“In terms of traditional mycotoxin detection, AI is not used yet. When it comes to collecting samples and analysing them for mycotoxin presence, there are several established techniques available. Instrumental methods such as LC- MS (Liquid Chromatography-Mass Spectrometry) are typically carried out in the lab, while faster on-site detection methods, such as dipstick tests or ELISA-based techniques, are more commonly used for quick results. These are the conventional methods used for detecting mycotoxins in samples. However, at our institute we are exploring the use of AI in combination with mass spectrometry. By analysing samples through targeted and untargeted screening, we gather extensive information about the compounds present and then AI is used to further identify these compounds.”
“I began with developing prediction models for mycotoxins back in 2006 when I started working at the Wageningen Food Safety Research Institute in the Netherlands. Initially, we used statistical methods, such as regression models, to predict mycotoxin presence in the field during harvest, based on climatic data and farming practices. Over time, we incorporated more mechanistic and biological models to simulate fungal infection rates and the presence of fungi. We then linked these biological models to mycotoxin presence using statistical methods. In the last 10 years, we have shifted to using machine learning, starting with Bayesian network modeling and expanding to other AI techniques. With the rapid advancements in AI, we continuously update our methods to see if they can improve our prediction models. Today, we use techniques like XGBoost and other AI models, sometimes in combination with biological models.”
“AI models offer more power than traditional statistical models, as they can better analyse the relationships between variables to predict mycotoxin presence. Additionally, AI is occasionally used to fill gaps in data. For example, in the Netherlands, we have around 30 weather stations from the Royal Netherlands Meteorological Institute (KNMI) that record temperature, rainfall, and humidity every 10 minutes. If there are missing data points, AI can interpolate values based on the surrounding data. One challenge with AI models people mention is the fear of the unknown and a lack of understanding about how they work. To address this, we use explainable AI at Wageningen Food Safety Research. Firstly, we rely on our own biological knowledge when developing models. Secondly, we use explainable AI to ensure that the important variables are correctly identified and understood in relation to mycotoxins. If any aspect of the model doesn’t align with biological expectations, we can adjust it accordingly. Regarding mass spectrometry in combination with AI, I see potential here, this approach allows for untargeted screening, where AI can then be used to identify compounds which can then be further analysed. This method not only applies to known mycotoxins but can help uncover more unknown metabolites.”
“When comparing the impact of weather and agronomic practices on mycotoxin presence, weather plays a much larger role. It has a greater influence on mycotoxin levels. While we cannot control the weather, we can adjust our practises, but the impact of optimising these practices is relatively low compared to the effects of weather. With agronomic practises, farmers can use models to predict key events, like flowering, and decide whether to apply fungicide spray. Models help target fungicide use more effectively, allowing farmers to spray only in areas or fields with high predicted mycotoxin levels. This makes fungicide use more sustainable. If, at a certain point the predictions all point to high mycotoxin levels, farmers can decide to switch to a more resistant crop variety or even consider planting a different crop species altogether.”
“Early detection, coupled with AI and advanced analytical methods, highlights the significant advantages of using AI in this context. This technology is crucial for early mycotoxin prediction, benefiting not only farmers but also buyers and food producers. If we can provide early predictions, like for example, where in Europe we foresee high and low mycotoxin levels, then risk managers and the supply chain have more time to act upon it. They can move from reactive to more proactive risk management. Buyers might choose to source from elsewhere, or repurpose the crop for other uses. While these models are just estimations, they are not the absolute truth, they may guide the sampling and analysis making monitoring more risk based, so that will also save costs in the end. I see great potential in using AI for this purpose.”
Join 13,000+ subscribers
Subscribe to our newsletter to stay updated about all the need-to-know content in the dairy sector, two times a week.
