The athlete microbiome project: Integrating deep learning to reveal microbial associations of physical fitness

Regular exercise improves human fitness and health through direct effects on muscle and metabolism and indirect effects involving the gut microbiome. To better understand the relationship between the gut microbiome and physical fitness, we conducted a secondary analysis of amplicon sequencing data and metadata from published human microbiota studies across three continents. Participants were categorized as athletes (n = 655) or non-athletes (n = 199) based on American Heart Association and American College of Cardiology guidelines. Using multivariate statistics, random forest, and a multilayer perceptron neural network, we identified significant differences in microbial community structure, diversity, and composition. Key taxa—including Faecalibacterium , Parabacteroides , and Prevotella —were associated with fat-free mass percentage and VO ₂ max, explaining up to 66% and 45% of the variance, respectively. The multilayer perceptron model distinguished athletes from non-athletes with 92.37% accuracy and an AUC of 0.97, highlighting distinct microbiome profiles between groups. These findings suggest microbial associations with athletic status and identify candidate taxa linked to physical fitness. Although causality cannot be inferred from this cross-sectional analysis, the results support further investigation into microbiome-mediated adaptations to exercise.