Machine learning provides preliminary snapshots of disease-specific variations in gut microbial aging-trajectories

Abstract

The human gut microbiome plays a critical role in metabolism, immunity, and disease resistance, with its composition shifting notably with age. These changes have positioned the microbiome as a potential “biological clock” capable of predicting host age. While numerous studies highlight strong associations between microbiome profiles and chronological age, most existing models are developed on healthy populations and rarely address how diseases alter microbial aging trajectories. Here, we compiled 19,342 gut microbiome samples from 67 study cohorts in 31 nations on both 16S and shotgun metagenomic sequencing. With Random Forest regression models trained on control samples only, we made robust age-microbiome correlations and computed interpretable species-level SHAP values to define aging signals. Test on disease cohorts showed systematic departures from healthy aging patterns, where numerous diseases started accelerated, decelerated, or disrupted microbial aging patterns. In 16 diseases, classification analyses also showed that aging-related features continue to distinguish diseased patients from controls, albeit with both cross-cohort consistencies and disease-specific variations. Our observations represent among the first large-scale demonstrations that microbiome-based indices of aging are replicable across populations but severely disrupted in disease. This research provides a foundation for the creation of disease-sensitive microbiome aging clocks and the discovery of microbial species responsible for healthy versus pathological aging trajectories with future biological interpretation and potential therapeutic implication.