OBJECTIVES: Increasing evidence indicates that the local microbiome can be used to predict host disease states. However, constructing models that obtain better results with fewer features is still challenging. METHODS: In this study, we established a nasal microbiome database consisting of 132 chronic rhinosinusitis patients, 27 nasal inverted papilloma patients, and 45 control patients. 16S rRNA gene sequencing was used to identify the species and abundance of bacteria in each sample, and a nasal microbiome database was generated after low-abundance bacteria were eliminated. The correlation data network of different groups of bacteria was constructed by calculating the correlation coefficient among bacterial genera, and the correlation parameters of the network were calculated based on graph theory. Through the development and application of a machine learning framework to optimize the screening process, combined with microbiome relationship network parameters based on graph theory, basic bacteria with high contributions to classification prediction were selected for the prediction of nasal diseases. RESULTS: We found that patients with nasal disease have a specific nasal microbiome signature and identified Moraxella, Prevotella, and Rothia as keystone genera that are markers of nasal disease
these markers can be interpreted as key control routes through graph theory analysis of the microbiota. With this strategy, we were able to characterize microbial community changes in nasal disease patients, which could reveal the potential role of the nasal microbiome in nasal disease. CONCLUSION: This study can provide a reference for the formulation of disease prevention and control policies. Our framework can be applied to other diseases to identify keystone genera that influence disease states and can be used to predict disease states.