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

Whether combining microbiome data from multiple body sites improves prediction, and whether different sites carry complementary or redundant information, are distinct questions that most studies conflate into a single accuracy metric. This work makes two contributions, one methodological and one biological, using paired stool and oral cavity microbiome samples from two independent cohorts. In the primary cohort (Ferretti et al., 2018; n=44), we show that a subject-matched fusion design combined with SHAP-based (SHapley Additive exPlanations) site attribution can detect complementary information between body sites even when no measurable accuracy gain results. Gut (stool) composition alone achieved near-perfect classification (area under the receiver operating characteristic curve, AUC = 1.00) of adult versus newborn age-group status, and combined stool-oral models never exceeded this ceiling. Despite the flat accuracy curve, SHAP analysis of the fused model showed that oral cavity features carried 58.1% of total feature importance versus 41.9% for stool, indicating real, non-redundant reliance on both sites. We replicated this finding in an independent, larger cohort (Brito et al., 2016; n=140) using a harder classification task (sex-based classification), where oral cavity again dominated the fused model’s SHAP importance (60.7% versus 39.3% for stool), and a permutation test confirmed the asymmetry was statistically significant (p < 0.005). The site-attribution asymmetry in the primary cohort was directionally consistent but did not reach significance, consistent with limited statistical power at n=44. The taxa driving these patterns include Malassezia restricta, Staphylococcus epidermidis , and Prevotella melaninogenica , all of which behave in a manner consistent with their established roles as early colonizers of the neonatal gut, skin, and oral cavity, once verified directly against abundance data. These results support a model of oral-gut microbiome maturation as two distinct, complementary processes, and demonstrate that detecting this relationship requires examining a model’s internal reasoning rather than its accuracy alone. The site-attribution method is released as a standalone, reusable Python tool.