Scientists Use AI To Accurately Predict If Listeria Will Survive Food Industry Disinfectants

LYNGBY, Denmark — Well-known disinfectants used in food inudstry settings might not be as foolproof as thought. New research shows scientists can now predict whether Listeria monocytogenes, a dangerous foodborne bacteria, will survive common quaternary ammonium disinfectants, using only its genetic code.

Thanks to the power of artificial intelligence, researchers achieved up to 97% accuracy in forecasting whether Listeria can tolerate the active ingredients found in many disinfectants.

Listeria monocytogenes causes listeriosis, a serious infection that can be fatal for pregnant women, newborns, elderly people, and those with compromised immune systems. The bacteria are so dangerous that the United States and Turkey have adopted “zero tolerance” policies for ready-to-eat products, meaning no detectable amounts are allowed in a 25-gram food sample.

The bacteria are particularly troublesome in their ability to persist in food processing environments for months or even years, forming tough biofilms and developing resistance to cleaning chemicals. Until now, determining whether a particular strain could survive disinfection required time-consuming laboratory tests. But researchers have discovered they can predict this survival with remarkable accuracy using only genetic information. Their paper is published in Scientific Reports.

On Tuesday, the FDA announced a Class I grade — the highest risk level — on a recall of 12,000 pounds of organic blueberries produced by Alma Pak International in Atlanta. The blueberries were said to have been contaminated with listeria, but shipped to just one customer in North Carolina.

How AI Predicts Listeria’s Disinfectant Tolerance

The research team at the Technical University of Denmark analyzed the complete genetic blueprints of 1,649 Listeria monocytogenes samples collected from North America and Europe. These bacteria came from food products (50%), food processing environments (30%), animals (5%), human clinical cases (5%), and farm environments (4%).

Each bacterial sample was tested against three different disinfectants: benzalkonium chloride and didecyldimethylammonium chloride (both quaternary ammonium compounds, which are chemicals commonly found in some household and many industrial cleaners), and Mida San 360 OM (a commercial disinfectant). The researchers then fed this information — genetic sequences paired with survival data — into machine learning algorithms to train artificial intelligence models.

The AI achieved balanced accuracy scores of up to 97% when predicting whether bacteria would survive benzalkonium chloride. For predicting the exact concentration needed to kill the bacteria, the models achieved mean squared errors as low as 0.07, indicating highly precise predictions.

Key Genes That Help Listeria Survive Cleaning Chemicals

The researchers used a technique called SHAP (which stands for Shapley Additive Explanations) to understand how the AI made its predictions. The models primarily focused on genes related to efflux pumps. These are molecular mechanisms that bacteria use to pump toxic substances out of their cells before they can cause damage.

The most important genetic features included known resistance genes like qacH and bcrC, which help bacteria survive quaternary ammonium compounds. But the AI also identified several previously unknown genes that might contribute to disinfectant tolerance, including genes related to cell wall structures and small DNA molecules called plasmids that can transfer between bacteria.

Manual analysis of the top genetic features revealed matches to various bacterial survival mechanisms, including transcriptional regulators (genes that control other genes), efflux transporters (pumps that remove toxins), transposases (enzymes that move genetic material), cell wall anchoring domains (structural proteins), and phage-related proteins (remnants of viruses that infect bacteria).

How Accurate Are These Predictions in Real-World Conditions?

While the AI models performed exceptionally well during training, testing them on independent datasets from three different research groups yielded more mixed results. The models achieved balanced accuracy scores ranging from 50% to 93%, with an overall performance of 67%.

This drop in performance exposes a crucial challenge in translating laboratory research to practical applications. Different research groups use varying experimental protocols, concentrations, and conditions when testing bacterial survival, making it difficult to create universally applicable prediction models.

The researchers also discovered that their AI classifier correctly identified the presence or absence of known resistance genes in almost all cases, with only one exception. When they tested their approach on two additional disinfectants, they found similarly promising results, with accuracy scores reaching 81% for didecyldimethylammonium chloride and 90% for the commercial disinfectant Mida San 360 OM.

What This Means for Food Processing and Food Safety

Food processing facilities spend enormous amounts of money on disinfection protocols, and knowing which bacteria are likely to survive could revolutionize how these operations design their cleaning strategies. Currently, food manufacturers often use a one-size-fits-all approach to disinfection, applying the same protocols regardless of which specific bacteria they’re dealing with.

This new predictive capability could enable more targeted and effective sanitation programs, potentially reducing food contamination incidents and associated costs. The researchers even discovered that models trained on pure compounds could predict survival against commercial disinfectants with 96% accuracy, indicating that the underlying genetic mechanisms are consistent across different product formulations.

Despite these promising results, the researchers acknowledge several important limitations. Most notably, their study examined bacteria grown in laboratory conditions as individual organisms, not as the complex biofilms that Listeria typically forms in real-world environments. Biofilms are communities of bacteria embedded in a protective matrix that makes them much more difficult to eliminate than individual cells.

Additionally, the study focused exclusively on quaternary ammonium compounds, which represent just one category of disinfectants used in food processing. The researchers suggest that their methodology could be extended to other disinfectant types, such as peracetic acid, for which resistance mechanisms are less well understood.

“AI does not provide us with a recipe for new disinfectants, but it does tell us which bacteria are likely to survive which chemicals. This enables swift and precise action,” says study co-author Pimlapas Shinny Leekitcharoenphon, a senior researcher at the DTU National Food Institute.

This research represents a significant step toward what the authors call “precision sanitation,” or the ability to tailor disinfection strategies based on the specific genetic makeup of bacterial contaminants. Just as precision medicine uses genetic information to customize treatments for individual patients, precision sanitation could customize cleaning protocols for specific bacterial threats.

As genome sequencing technology becomes faster and more affordable, food processors could sequence bacterial contaminants and predict their disinfectant tolerance within hours rather than days. In a world where foodborne illness affects millions of people annually, the ability to predict and prevent bacterial survival could save countless lives and transform how we think about food safety.