Machine Learning Reveals Microbial Taxa Associated with a Swim across the Pacific Ocean.

Abstract

Purpose: This study aimed to characterize the association between microbial dynamics and excessive exercise. Methods: Swabbed fecal samples, body composition (percent body fat), and swimming logs were collected (n = 94) from a single individual over 107 days as he swam across the Pacific Ocean. The V4 region of the 16S rRNA gene was sequenced, generating 6.2 million amplicon sequence variants. Multivariate analysis was used to analyze the microbial community structure, and machine learning (random forest) was used to model the microbial dynamics over time using R statistical programming. Results: Our findings show a significant reduction in percent fat mass (Pearson; p < 0.01, R = -0.89) and daily swim distance (Spearman; p < 0.01, R = -0.30). Furthermore, the microbial community structure became increasingly similar over time (PERMANOVA; p < 0.01, R = -0.27). Decision-based modeling (random forest) revealed the genera Alistipes, Anaerostipes, Bifidobacterium, Butyricimonas, Lachnospira, Lachnobacterium, and Ruminococcus as important microbial biomarkers of excessive exercise for explaining variations observed throughout the swim (OOB; R = 0.893). Conclusions: We show that microbial community structure and composition accurately classify outcomes of excessive exercise in relation to body composition, blood pressure, and daily swim distance. More importantly, microbial dynamics reveal the microbial taxa significantly associated with increased exercise volume, highlighting specific microbes responsive to excessive swimming.