Scientists Use Machine Learning To Help Fight Antibiotic Resistance In Farmed Chickens

Scientists have used machine learning to find new ways to identify and pinpoint disease in poultry farms, which will help to reduce the need for antibiotic treatment, lowering the risk of antibiotic resistance transferring to human populations.

In this project, researchers in Nottingham collected samples from the animals, humans and environment on a Chinese farm and its connected slaughterhouse. This complex "big" data has now been analyzed for new diagnostic biomarkers that will predict and detect bacterial infection, insurgence of AMR, and transfer to humans. This data will then allow early intervention and treatment, reducing spread and the need for antibiotics.

The study produced three key findings. Firstly, several similar clinically relevant antimicrobial resistance genes (ARGs) and associated mobile genetic elements (antibiotic resistance genes able to move within genomes and between bacteria), were found in both human and broiler chicken samples. In particular, eleven types of clinically important antibiotic resistance genes, with conserved mobile ARG gene structures were found between samples from different hosts.

Dr. Dottorini said, "These similarities would have been missed if we only used large-scale conventional comparative analysis, which in fact showed that microbiome and resistomes differ across environments and hosts. Overall, this finding suggests the relevance of adopting a multi-scale analysis when dissecting similarities and differences of resistomes and microbiomes in complex interconnected environments."

Secondly, the study showed that by developing a machine learning-powered approach integrating metagenomics data with culture-based methods, the team found the existence of a core chicken gut resistome that is correlated with the AMR circulating in the farms. These results supported the hypothesis that correlations exist between resistance phenotypes of individual commensal and pathogenic bacteria and the types of ARGs in the resistome in which they exist.

Finally, using sensing technology and machine learning, the team uncovered that the AMR-related core resistomes are themselves associated with various external factors such as temperature and humidity.