Tongue coating microbiota-based machine learning for diagnosing digestive system tumours.

Abstract

Background: Digestive system tumours (DSTs) often diagnosed late due to nonspecific symptoms. Non-invasive biomarkers are crucial for early detection and improved outcomes.

Patients and methods: We collected tongue coating samples from 710 patients diagnosed with DST and 489 healthy controls (HC) from April 2023, to December 2023. Microbial composition was analyzed using 16S rRNA sequencing, and five machine learning algorithms were applied to assess the diagnostic potential of tongue coating microbiota.

Results: Alpha diversity analysis showed that the microbial diversity in the tongue coating was significantly increased in DST patients. LEfSe analysis identified DST-enriched genera Alloprevotella and Prevotella, contrasting with HC-dominant taxa Neisseria, Haemophilus, and Porphyromonas (LDA >4). Notably, when comparing each of the four DST subtypes with the HC group, the proportion of Haemophilus in the HC group was significantly higher, and it was identified as an important feature for distinguishing the HC group. Machine learning validation demonstrated superior diagnostic performance of the Extreme Gradient Boosting (XGBoost) model, achieving an AUC of 0.926 (95% CI: 0.893-0.958) in internal validation, outperforming the other four machine learning models.

Conclusion: Tongue coating microbiota shows promise as a non-invasive biomarker for DST diagnosis, supported by robust machine learning models.