Hypothesis

Microbiome-derived metabolites induce cohort-specific shifts in the epigenetic manifold that accelerate biological age; aligning these manifolds across populations using optimal transport uncovers a causal microbiome‑epigenome axis that outperforms static multi-omics clocks in mortality prediction.

Mechanistic Rationale

Recent multi-omics clocks show that microbiome age acceleration uniquely associates with numerous physiological parameters, yet most models treat each omics layer as additive features within a shared embedding space [1]. This ignores the fact that microbial metabolites (e.g., short‑chain fatty acids, secondary bile acids) can directly modulate host DNA methyltransferases and histone acetyltransferases, reshaping the epigenetic landscape in a manner that varies with diet, geography, and host genetics [2]. Consequently, the true biological age signal resides in a low‑dimensional manifold that is differentially warped across cohorts. Current methods that concatenate omics or apply simple linear alignments (e.g., PCA, PLS) fail to capture these nonlinear warps, leading to only marginal gains when adding new modalities [3].

Optimal transport (OT) provides a principled way to compute a mapping between probability distributions that minimizes a cost function, effectively "warping" one manifold onto another while preserving intrinsic geometry [4]. By learning an OT map that aligns the joint transcript‑prote‑metabo‑epi‑microbiome distributions of a reference cohort to those of a target cohort, we can isolate the residual variation attributable to microbiome‑driven epigenetic remodeling.

Testable Predictions

1. Manifold Alignment Improves Mortality Stratification – A model that first applies OT‑based alignment to multi-omics data from diverse cohorts (e.g., UKBB, Framingham, and a non‑Western population) and then trains a LightGBM clock on the aligned embeddings will achieve significantly lower RMSE and higher C‑index for all‑cause mortality than a clock trained on raw concatenated features (p<0.01, DeLong test).
2. Microbiome‑Specific Transport Cost Predicts Epigenetic Age Acceleration – The OT cost contributed specifically by the microbiome modality (i.e., the Wasserstein distance between microbiome distributions) will correlate positively with epigenetic age acceleration residuals (measured as DNAmAge − chronologicalAge) across individuals (Spearman ρ > 0.3, FDR < 0.05).
3. Interventional Shift in Microbiome Aligns Manifolds – In a controlled dietary intervention (high‑fiber vs. low‑fiber) within a single cohort, the OT map derived from pre‑ to post‑intervention microbiome profiles will predict changes in DNAmAge at follow‑up (R² > 0.2). Failure to observe this relationship falsifies the microbiome‑driven manifold warp hypothesis.

Experimental Design

• Data: Multi-omics datasets (transcriptome, proteome, metabolome, epigenome, 16S/metagenome) from at least three geographically distinct cohorts with longitudinal mortality follow‑up (minimum 5 years).
• Alignment: Compute pairwise OT maps between each cohort’s joint omics distribution and a reference distribution using the Sinkhorn algorithm with entropy regularization; retain the microbiome‑specific transport plan.
• Modeling: Train mortality risk models (LightGBM and Cox PH) on (a) raw concatenated features, (b) features after generic manifold alignment (e.g., CCA), and (c) features after microbiome‑guided OT alignment.
• Validation: Nested cross‑validation within each cohort and external hold‑out cohorts; assess calibration (Hosmer‑Lemeshow) and discrimination (time‑dependent AUC).
• Falsification: If OT‑aligned models do not outperform baseline models by a statistically significant margin, or if microbiome transport cost fails to correlate with epigenetic acceleration, the hypothesis is refuted.

Implications

Demonstrating that microbiome‑induced manifold warps are a core driver of epigenetic aging would shift focus from static multi‑omics feature integration to dynamic, transport‑based alignment strategies. It would also suggest that microbiome‑targeted interventions could reset biological age trajectories, offering a mechanistic bridge between multi‑omics prediction and actionable gerontology.