Abstract
BACKGROUND: The application of clinical mNGS for diagnosing respiratory infections improves etiology diagnosis, however at the same time, it brings new challenges as an unbiased sequencing method informing all identified microbiomes in the specimen. METHODS: Strategy evaluation and metagenomic analysis were performed for the mNGS data generated between March 2017 and October 2019. Diagnostic strengths of four specimen types were assessed to pinpoint the more appropriate type for mNGS diagnosis of respiratory infections. Microbiome complexity was revealed between patient cohorts and infection types. A bioinformatic pipeline resembling diagnosis results was built based upon multiple bioinformatic parameters. RESULTS: The positive predictive values (PPVs) for mNGS diagnosing of non-mycobacterium, Nontuberculous Mycobacteria (NTM), and Aspergillus were obviously higher in bronchoalveolar lavage fluid (BALF) demonstrating the potency of BALF in mNGS diagnosis. Lung tissues and sputum were acceptable for diagnosis of the Mycobacterium tuberculosis (MTB) infections. Interestingly, significant taxonomy differences were identified in sufficient BALF specimens, and unique bacteriome and virome compositions were found in the BALF specimens of tumor patients. Our pipeline showed comparative diagnostic strength with the clinical microbiological diagnosis. CONCLUSIONS: To achieve reliable mNGS diagnosis result, BALF specimens for suspicious common infections, and lung tissues and sputum for doubtful MTB infections are recommended to avoid the false results given by the complexed respiratory microbiomes. Our developed bioinformatic pipeline successful helps mNGS data interpretation and reduces manual corrections for etiology diagnosis.