Machine-learning algorithms for asthma, COPD, and lung cancer risk assessment using circulating microbial extracellular vesicle data and their application to assess dietary effects.

Although mounting evidence suggests that the microbiome has a tremendous influence on intractable disease, the relationship between circulating microbial extracellular vesicles (EVs) and respiratory disease remains unexplored. Here, we developed predictive diagnostic models for COPD, asthma, and lung cancer by applying machine learning to microbial EV metagenomes isolated from patient serum and coded by their accumulated taxonomic hierarchy. All models demonstrated high predictive strength with mean AUC values ranging from 0.93 to 0.99 with various important features at the genus and phylum levels. Application of the clinical models in mice showed that various foods reduced high-fat diet-associated asthma and lung cancer risk, while COPD was minimally affected. In conclusion, this study offers a novel methodology for respiratory disease prediction and highlights the utility of serum microbial EVs as data-rich features for noninvasive diagnosis.