Interpretable Machine Learning Reveals Complementary Age-Related Signatures in the Oral and Gut Microbiome

A pioneering study examines whether combining microbiome samples from the oral cavity and gut provides complementary insights into age-related changes, rather than simply enhancing predictive accuracy. Researchers analyzed paired stool and oral microbiome samples from 44 individuals, both healthy adults and newborns, to uncover novel biological patterns using advanced analytical techniques.

The study's methodological innovation lies in the subject-matched fusion design combined with SHAP-based site attribution. This approach enabled the detection of complementary information between body sites, even when no measurable improvement in predictive accuracy was observed. Conventional model comparison methods often misread this pattern as a null result, but the researchers confirmed that this flat accuracy curve actually reflects a genuine biological signal.

Interestingly, despite the lack of accuracy gain, the oral cavity features contributed more to the model's total feature importance (58.1%) compared to stool features (41.9%). This finding suggests that the model is extracting meaningful, non-redundant information from both body sites. The taxa driving this complementary pattern include Malassezia restricta, Staphylococcus epidermidis, and Prevotella melaninogenica.

These organisms align with their established roles as early colonizers of the neonatal gut, skin, and oral cavity, verified through direct comparison with abundance data rather than relying on literature inference.

An independent, larger study using a different analytical method reported a compatible pattern, reinforcing the findings. Together, these results suggest that the oral-gut microbiome maturation process comprises two distinct, complementary processes. The study emphasizes the importance of examining a model's internal reasoning rather than solely relying on its accuracy to uncover such complex biological relationships.