Exploring fecal microbiota signatures associated with immune response and antibiotic impact in NSCLC: insights from metagenomic and machine learning approaches.

Background: Substantial interstudy heterogeneity in cancer immunotherapy-associated biomarkers has hindered their clinical applicability. To address this challenge, we performed a comprehensive integration of publicly available global metagenomic datasets. By leveraging metagenomic profiling and machine learning approaches, this study aimed to elucidate gut microbial signatures associated with immune response in lung cancer (LC) and to evaluate the modulatory effects of antibiotic exposure. Methods: A systematic literature search was conducted to identify relevant datasets, resulting in the inclusion of 209 fecal metagenomic samples: 154 baseline samples (45 responders, 37 non-responders, and 72 healthy controls) and 55 longitudinal samples collected during immunotherapy. We performed taxonomic and functional characterization of gut microbiota (GM) differentiating responders from non-responders, delineated microbiome dynamics during treatment, and assessed the impact of antibiotics on key microbial taxa. Among eight machine learning algorithms evaluated, the optimal model was selected to construct a predictive framework for immunotherapy response. Results: Microbial α-diversity was significantly elevated in responders compared to non-responders, with antibiotic administration further amplifying this difference-most notably at the species level. Integrative multi-omics analysis identified two pivotal microbial biomarkers, Conclusions: This study represents the first large-scale, cross-cohort integration of metagenomic data to identify reproducible GM signatures predictive of immune checkpoint inhibitor efficacy in LC. The findings not only underscore the prognostic relevance of specific taxa but also establish a foundation for developing microbiome-informed, personalized immunotherapeutic strategies.