Metagenomic Analysis of Bronchoalveolar Lavage Fluid Enables Differential Diagnosis Between Lung Cancer and Pulmonary Infections

Abstract

Recent advances in unbiased metagenomic next-generation sequencing (mNGS) have enabled the simultaneous examination of both microbial and host genetic material in a single test. This study harnesses cost-effective bronchoalveolar lavage fluid (BALF) mNGS data from patients with lung cancer (n=123) and pulmonary infections (n=279). We developed a machine learning-based diagnostic approach to differentiate between these two conditions, which are often misdiagnosed in clinical settings. To ensure independence between model construction and validation, we divided the cohorts based on the collection dates of the samples. The training cohort (lung cancer, n=87; pulmonary infection, n=197) revealed distinct differences in DNA/RNA microbial composition, bacteriophage abundances, and host responses, including gene expression, transposable element levels, immune cell composition, and tumor fraction determined by copy number variation (CNV). These features, blinded to the validation cohort, were integrated into a host/microbe metagenomics-driven machine learning model (Model VI). The model demonstrated an Area Under the Curve (AUC) of 0.87 (95% CI = 0.857-0.883) in the training cohort and 0.831 (95% CI = 0.819-0.843) in the validation cohort for differentiating between patients with lung cancer and pulmonary infections. Applying a composite predictive model based on a rule-in and rule-out strategy significantly increased accuracy in distinguishing lung cancer from tuberculosis (ACC=0.913), fungal infection (ACC=0.955), and bacterial infection (ACC=0.836). These results underscore the potential of mNGS-based analysis as a valuable, cost-effective tool for the early differentiation of lung cancer from pulmonary infections, offering a comprehensive testing solution in a clinical context.