CARD:Epi - Contextualizing Antimicrobial Resistance Determinants Using Deep Learning Language Models

The spread of infections caused by drug-resistant bacteria is a growing concern in healthcare, agriculture, and livestock. Researchers have identified several key factors that contribute to the uncontrolled spread of these infections, including the environment, pathogens, hosts, and antimicrobial resistance genes (ARGs). Each individual study sheds light on the devastating impact of drug-resistant infections, but when combined, they offer important contextual information about the transmission of ARGs.

To analyze this data, the researchers utilized a biomedical deep-learning language model called BioBERT. BioBERT was trained on two tasks: entity recognition, which identified AMR-relevant terms such as ARGs, taxonomy, environments, and geographical locations, and relation extraction, which determined how these terms contextualized ARGs. By analyzing results from 204,094 antimicrobial resistance publications worldwide, the team generated interpretable results about the sources where genes are commonly found.

To further visualize the dataset, two pipelines were created to analyze transmission patterns of ARGs across agriculture, environments, and human populations. These pipelines used a Confusogram and Uniform Manifold Approximation and Projection to provide a comprehensive understanding of how scientific literature can be used to assess transmission patterns of ARGs. This large-scale approach collects antimicrobial resistance data from a lesser-known resource - the systematic examination of the vast body of AMR literature.