Scientists Predict Listeria Resistance to Disinfectants with AI
Scientists have achieved up to 97% accuracy in predicting which strains of Listeria monocytogenes can survive common disinfectants by analyzing their DNA using artificial intelligence. The AI not only identified known resistance genes but also uncovered previously unknown genetic factors involved in bacterial survival. This insight suggests that resistance arises through multiple genetic pathways.
This advance could lead to “precision sanitation” in food safety, where cleaning protocols are customized based on the genetic makeup of bacterial contaminants.
Introduction
Common disinfectants used in food processing may not be as effective as once believed. Researchers from the Technical University of Denmark have demonstrated that the survival of Listeria monocytogenes against quaternary ammonium disinfectants can be predicted solely from genetic data, using machine learning models.
Listeria monocytogenes causes listeriosis, a severe infection posing high risk to pregnant women, newborns, elderly, and immunocompromised individuals. Due to its danger, “zero tolerance” policies exist in countries like the United States and Turkey for ready-to-eat foods, allowing no detectable bacteria in a 25-gram sample.
The bacteria's ability to persist in food processing facilities for months or years, often forming biofilms and developing disinfectant resistance, complicates control efforts. Until now, confirming resistance involved time-intensive lab testing. This new AI-based approach can predict survival quickly and accurately using genetic information alone. Their findings were published in Scientific Reports.
In a related event, the FDA recently issued a Class I recall on 12,000 pounds of organic blueberries contaminated with listeria, underscoring the ongoing food safety challenges.
How AI Predicts Listeria’s Disinfectant Tolerance
The research team analyzed whole-genome sequences of 1,649 Listeria monocytogenes isolates collected across North America and Europe. These samples originated from food products (50%), food processing environments (30%), animals (5%), human clinical cases (5%), and farm environments (4%).
Each isolate was tested against three disinfectants: benzalkonium chloride and didecyldimethylammonium chloride (both quaternary ammonium compounds), and Mida San 360 OM, a commercial disinfectant.
The researchers combined genetic data with survival results to train machine learning models. The AI achieved balanced accuracy rates up to 97% when predicting survival against benzalkonium chloride. It also precisely predicted minimum inhibitory concentrations with mean squared errors as low as 0.07.
Key Genes That Help Listeria Survive Cleaning Chemicals
To interpret the AI’s decision-making, the team used SHAP (Shapley Additive Explanations), revealing that genes encoding efflux pumps were central. Efflux pumps expel toxic substances from bacterial cells, enhancing survival.
Important genes identified included known resistance genes like qacH and bcrC. Additionally, the AI discovered several previously unrecognized genes potentially contributing to resistance. These involved genes related to cell wall structures and plasmids—small DNA molecules that can transfer between bacteria.
Manual examination of top genetic features matched mechanisms such as transcriptional regulators, efflux transporters, transposases, cell wall anchoring domains, and phage-related proteins, all linked to bacterial survival.
Accuracy of Predictions in Real-World Conditions
When tested on independent datasets from three other research groups, prediction accuracy varied between 50% and 93%, averaging 67%. This reduction highlights challenges in applying lab-developed models to practical conditions, where experimental protocols and testing environments differ significantly.
Despite this, the AI correctly identified the presence or absence of known resistance genes in nearly all cases. Testing on two additional disinfectants yielded promising accuracy scores of 81% and 90%.
Implications for Food Processing and Safety
Food processing industries invest heavily in disinfection. Understanding which bacterial strains survive specific disinfectants can help optimize cleaning protocols, moving away from uniform approaches to targeted sanitation strategies.
Interestingly, models trained on pure disinfectant compounds accurately predicted survival against commercial disinfectant formulations, suggesting consistent underlying genetic resistance mechanisms.
However, the study has limitations. It focused on bacteria grown individually under lab conditions, while in real settings, Listeria often exists within biofilms that are harder to eradicate. Also, only quaternary ammonium compounds were examined, leaving out other disinfectant classes like peracetic acid.
One of the researchers noted, “AI doesn’t provide new disinfectant formulas but identifies which bacteria are likely to survive specific chemicals, enabling faster and more precise interventions.” This approach marks a step toward precision sanitation—designing disinfection strategies informed by genetic data, akin to precision medicine in healthcare.
As genome sequencing becomes faster and more affordable, food processors could soon identify bacterial resistance profiles within hours, improving food safety and reducing illness risks.
Paper Summary
Methodology
- Analyzed 1,649 Listeria monocytogenes samples from North America and Europe using whole genome sequencing.
- Tested bacterial survival against three disinfectants.
- Trained machine learning models with genetic and survival data.
- Applied a phylogeny-aware approach to group genetically similar bacteria during model training and testing.
Results
- Achieved balanced accuracy up to 97% for predicting survival against benzalkonium chloride.
- Mean squared error as low as 0.07 for predicting minimum inhibitory concentrations.
- Identified known and novel genetic features linked to disinfectant tolerance.
- Independent datasets showed variable accuracy (50-93%).
Limitations
- Laboratory-grown individual bacteria were studied, not biofilms.
- Focus on quaternary ammonium compounds limits generalizability.
- Variability across datasets suggests need for standardized testing protocols.
- Geographic restriction of samples may affect applicability elsewhere.
Funding and Disclosures
The research received funding from Karl Pedersen og Hustrus Industrifond, the Danish Dairy Research Foundation, the Milk Levy Fund, and the MRC Centre for Global Infectious Disease Analysis. The authors declared no competing interests.
Publication Information
This study was published in Scientific Reports in 2025 (Volume 15, Article 10382) by researchers from the Technical University of Denmark and Imperial College London.
For more information on machine learning applications in microbiology and AI in research, visit the Complete AI Training latest courses.
Your membership also unlocks:
