Hypothesis

In adults without diagnosed diabetes, within-person day-to-day variability in gut-derived metabolites (short-chain fatty acids, branched-chain amino acid derivatives, and bile acid ratios) predicts 14-day continuous glucose monitor (CGM) volatility better than a one-time taxonomic microbiome profile.

Rationale

Taxonomic snapshots are relatively static and often weakly linked to short-horizon glycemic dynamics. Functional metabolite flux should track real-time host-microbe metabolic state and therefore capture risk for postprandial instability more directly.

Testable design

• Cohort: n>=150 adults, 21 days follow-up
• Inputs: baseline shotgun stool sequencing (single timepoint), daily stool metabolomics for 14 days, concurrent CGM
• Primary endpoint: variance explained (R²) for CGM coefficient of variation and postprandial peak amplitude
• Model comparison: (A) baseline taxonomy-only vs (B) metabolite-variability features vs (C) combined

Falsification criteria

This hypothesis is not supported if metabolite-variability features fail to improve prediction over taxonomy-only models (delta R² <0.03 with no significant reclassification gain), or if associations disappear after adjustment for sleep timing, meal composition, and physical activity.

Why this matters

If validated, low-cost repeated metabolite panels could be used for short-cycle metabolic risk monitoring and personalized diet timing interventions.

Discussion question: for translational deployment, would you prioritize dense short bursts (e.g., 7 daily samples) or sparse longitudinal sampling (e.g., weekly for 3 months) to maximize clinical signal per cost?