Gut microbiota signatures and machine learning-based candidate feature prioritization in advanced colorectal cancer.

This single-center, cross-sectional case-control study aimed to characterize the gut microbiome profiles of patients with advanced colorectal cancer (CRC), identify candidate microbial features, and explore how well these features distinguished advanced CRC cases from healthy controls within the study dataset. Fecal samples were collected from 72 treatment-naïve patients with advanced CRC and 61 healthy controls. Microbial community structure was profiled using high-throughput sequencing of the V3-V4 region of the 16 S rRNA gene. The analytical pipeline included α/β-diversity analysis, multilevel taxonomic analysis, LEfSe, and machine-learning algorithms, including random forest (RF), gradient boosting machine (GBM), and LASSO, to identify key Amplicon Sequence Variants (ASVs) and evaluate their ability to discriminate between the two groups. Results showed significantly reduced α-diversity and distinct β-diversity in the CRC group. Key SCFA-producing genera (Faecalibacterium, Agathobacter, Roseburia) were consistently depleted. The RF feature-prioritization model achieved an out-of-bag (OOB) accuracy of 0.850 and an OOB AUC of 0.899. Nested five-fold cross-validation based on the prioritized ASVs yielded AUCs of 0.891, 0.900, and 0.891 for RF, GBM, and LASSO, respectively. These findings show internally reproducible case-control discriminatory patterns within the present cohort and support further evaluation of the prioritized microbial features in independent, clinically representative populations.