By merging genetic sequencing and the analysis of microbes present in a milk sample with artificial intelligence (AI), scientists have discovered abnormalities in milk production such as contamination or the presence of unauthorized additives. This innovative method could enhance safety in the dairy industry.
Through the integration of genetic sequencing and microbial analysis from milk samples with artificial intelligence (AI), researchers have successfully identified discrepancies in milk production, including contamination and unapproved additives. This pioneering method may boost dairy safety, as indicated by the study’s authors representing Penn State, Cornell University, and IBM Research.
According to findings published in mSystems, a journal associated with the American Society for Microbiology, the team utilized shotgun metagenomics data alongside AI to identify antibiotic-treated milk that was secretly added to bulk tank samples they collected for analysis. To confirm their results, the researchers employed their explainable AI tool on publicly available genetically sequenced datasets of bulk milk, thereby showcasing the reliability of their untargeted approach.
Erika Ganda, the lead author and assistant professor focused on food animal microbiomes at Penn State College of Agricultural Sciences, remarked, “This was a proof of concept study. By analyzing data from the microbes in raw milk with the aid of artificial intelligence, we can determine characteristics such as whether the milk is pre-pasteurization, post-pasteurization, or sourced from a cow treated with antibiotics.”
The researchers gathered 58 bulk tank milk samples and implemented various AI algorithms to distinguish baseline samples from those indicating potential anomalies, such as milk from external farms or milk containing antibiotics. This research provided deeper insights into raw milk metagenomes—genomic collections from numerous individual microbes in a sample—beyond what has been published so far, confirming the presence of stable microbial elements across samples.
The findings of this study indicate that AI holds substantial promise for enhancing the identification of abnormalities in food production, offering a comprehensive technique to bolster scientists’ efforts in ensuring food safety, according to Ganda.
“Conventional methods for analyzing microbial sequencing data, such as assessing alpha and beta diversity and clustering, were not as proficient in distinguishing between normal and anomalous samples,” she explained. “The integration of AI, however, enabled precise classification and identification of microbial influences related to anomalies.”
Kristen Beck, the first author of the study and a senior research scientist at IBM Research, noted that microbial systems and the food supply chain represent optimal candidates for AI application due to the complex and dynamic interactions among microbes.
“Numerous factors within the food supply chain can impact the signals we aim to detect,” she added. “AI assists in separating relevant signals from background noise.”
While this research centers on dairy production, its implications extend to the larger food industry. Ganda highlighted that milk was chosen as a model due to its status as a primary ingredient for fluid milk, a high-demand product where there are serious concerns about fraud, especially in developing countries.
Challenges concerning food quality and safety can trigger significant repercussions along the supply chain, resulting in considerable health and economic consequences, Ganda pointed out. Thus, there is a keen interest in employing both targeted and untargeted strategies to detect ingredients or food products that are at higher risk for fraud, quality degradation, or safety issues.
“Untargeted approaches describe all identifiable molecules to recognize ingredients or products that diverge from a ‘baseline state’ deemed normal or controlled,” she noted. “Crucially, these untargeted methods function as screening tools that do not categorically label an ingredient or product as unsafe or adulterated but rather indicate a departure from the norm that may necessitate further investigation.”
This unique collaboration maximized the strengths of each partner involved, with Ganda emphasizing the use of IBM’s open-source AI technology, Automated Explainable AI for Omics, which enabled the processing of extensive metagenomic data or nucleotide sequences from microbes in bulk milk. This capability allows for the identification of microbial signatures that conventional techniques might overlook. The expertise from Cornell in dairy science enriched the study’s practical relevance for the dairy sector, while Penn State’s One Health Microbiome Center contributed significantly to the integration of microbial data for broader health and safety applications.
Other contributors to the research included Niina Haiminen, Akshay Agarwal, Anna Paola Carrieri, Matthew Madgwick, Jennifer Kelly, and Ban Kawas from IBM Research; Victor Pylro from the Federal University of Lavras, Brazil; and Martin Wiedmann from Cornell University.
The U.S. Department of Agriculture provided support for this research through Penn State.
