Integration of Multiomic and Multi-phenotypic Data Identifies Biological Pathways Associated with Physical Fitness

Azar Alizadeh, John Graf, Matthew J. Misner, Andrew Burns, Fiona Ginty, Kevin J. O’Donovan, J. Kenneth Wickiser, N Barringer, Gregory M Freisinger, Neil Hermansen, J.e. Mcdonough, Brian M. Davis, Evelina Loghin, Christine Surrette, Peter Tu, Justin Welch, Oliver Boomhower, Ralf Lenigk, Rachel Sorell, Tyler C. Hammond, Sara Peterson, Alexandre Caron, Leila Safa, Chrystal Chadwick, Stephanie Stacey, J. Jobin, Scott Evans, Rui Xue, Gurvinder Singh Khinda, Eric Williams, Swapnil R. Chhabra, Nhan Huynh, Ernest Fraenkel, Luca Marinelli

Abstract: Unraveling the complex associations between human phenotypes and molecular pathways can pave the way to improved health and performance, but faces a fundamental challenge: the measurable genes, proteins, and metabolites vastly outnumber the participants in even the largest studies, yielding spurious correlations. To address this imbalance, we have developed a bioinformatic framework and computational approach (“PhenoMol”) to discover biological drivers of phenotypic characteristics that integrates all available phenotypic data predictive of outcomes and reduces multi-omic data dimensionality by generating “expression circuits” via graph theory constrained by prior biological knowledge of molecular interactions. We applied PhenoMol to analyze causal patterns and predict elite physical performance in a healthy cohort with deep physiological, physical, behavioral, cognitive, and molecular characterization. PhenoMol outperforms regression models based on equivalent analytic methodologies that do not employ network biology for dimensionality reduction. The PhenoMol software is provided for future studies.

Introduction: The "Small N, Large P" Challenge

In the era of "big data" biology, researchers can measure tens of thousands of molecular features—genes, proteins, and metabolites—from a single blood draw. However, high-performance human studies often involve relatively small cohorts, typically dozens or hundreds of participants. This creates a fundamental statistical hurdle known as the "small $N$, large $P$" problem: the number of variables ($P$) vastly outnumbers the participants ($N$), leading to spurious correlations and models that fail to generalize to broader populations.

Standard statistical methods often struggle to distinguish true biological drivers from background noise. There is a critical need for frameworks that can integrate diverse data types—ranging from psychological traits and wearable sensor data to deep molecular profiling—to identify the actual mechanistic pathways associated with physical fitness and "biological aptitude."

The PhenoMol Framework

To address these challenges, we introduced PhenoMol, a bioinformatic pipeline that utilizes network biology as a filter to move beyond simple correlations. The framework operates through several key stages:

  1. Interactome Constraints: Rather than treating every molecule as an independent variable, PhenoMol utilizes a map of known human molecular interactions (protein-protein and protein-metabolite).
  2. Prize-Collecting Steiner Forest (PCSF): This graph-theory algorithm identifies "Principal Networks" (PNs). It selects molecules that correlate strongly with performance ("prizes") only if they are mechanistically linked to other high-value molecules, pruning the search space from ~50,000 features down to approximately 120 robust nodes.
  3. Louvain Clustering: The resulting network is partitioned into functional modules, or "expression circuits."
  4. Ensemble Modeling: These modules are used to build predictive models via Sparse Partial Least Squares Regression (sPLSR).
Figure 1: Overview of the study design, including ACFT events (A), phenotypic and molecular data layers (B), and performance distributions across the cohort (C-H).

This approach ensures that the features selected for the final model are not just statistically significant but are also biologically plausible.

Study Cohort and Data Integration

The study utilized a unique, longitudinal dataset from the Measuring Biological Aptitude (MBA) program, involving 86 cadets (71 male, 15 female) from the U.S. Military Academy at West Point. The dataset is exceptionally deep, encompassing approximately 140 phenotypic features and over 50,000 molecular features.

Table 1: Phenotypic and performance characteristics of the study cohort, highlighting significant physiological and fitness differences between male and female participants.

Phenotypic data included Army Combat Fitness Test (ACFT) scores, VO2max, body composition, personality assessments (NEO PI-3), cognitive tests, and continuous tracking via Oura rings and Garmin watches. Multi-omic data included DNA methylation, mRNA sequencing, untargeted proteomics, and metabolomics. Participants underwent a standardized stress protocol involving intense cycling bouts (~70% VO2max) with blood draws at baseline and multiple post-exercise intervals.

Biological Drivers of Elite Performance

While phenotypic models (using metrics like body fat percentage or VO2max) can predict fitness, PhenoMol identifies the underlying molecular machinery, bridging the gap between high performance and the cellular processes enabling it. Key biological insights include:

  • Bile Acids and Muscle: A strong association was found between primary bile acid biosynthesis and high performance. While typically associated with digestion, emerging evidence suggests bile acids act as signaling molecules for skeletal muscle mass and energy metabolism.
  • The Complement System: The complement and coagulation cascades, usually linked to immunity, were identified as top predictors. This suggests elite performers may have more efficient systemic responses to the "injury" and inflammation caused by strenuous exercise, facilitating faster recovery.
  • Metabolic Efficiency: High performers exhibited distinct signatures in the urea cycle and arginine/proline metabolism, pathways critical for clearing toxic ammonia and producing nitric oxide to improve blood flow.

Results and Validation

PhenoMol models significantly outperformed standard regression "control" models that lacked network constraints, achieving a lower Root Mean Square Error ($p = 0.00011$). Permutation testing confirmed the model's accuracy was not due to chance ($p = 0.0033$). Furthermore, the model successfully differentiated "elite" performers (ACFT score > 540) using molecular "expression circuits," integrating disparate data—such as the long non-coding RNA JPX and the metabolite ascorbic acid 3-sulfate—into a single, coherent predictive network.

Limitations and Future Directions

A primary limitation of this study is the gender imbalance; the small number of female participants (N=14 with full data) prevented the application of the full PhenoMol pipeline to the female cohort separately. Future research should focus on larger, elite female cohorts to identify gender-specific biological markers. Additionally, the West Point cohort represents an exceptionally fit population, which may bias the model toward markers of elite performance rather than general health.

The identified "expression circuits" could eventually serve as targets for personalized nutrition, supplements, or training regimens to enhance resilience. To support further research, the PhenoMol software has been made open-source.

Summary updated

Abstract

ABSTRACT Unraveling the complex associations between human phenotypes and molecular pathways can pave the way to improved health and performance, but faces a fundamental challenge: the measurable genes, proteins, and metabolites vastly outnumber the participants in even the largest studies, yielding spurious correlations. To address this imbalance, we have developed a bioinformatic framework and computational approach (“PhenoMol”) to discover biological drivers of phenotypic characteristics that integrates all available phenotypic data predictive of outcomes and reduces multi-omic data dimensionality by generating “expression circuits” via graph theory constrained by prior biological knowledge of molecular interactions. We applied PhenoMol to analyze causal patterns and predict elite physical performance in a healthy cohort with deep physiological, physical, behavioral, cognitive, and molecular characterization. PhenoMol outperforms regression models based on equivalent analytic methodologies that do not employ network biology for dimensionality reduction. The PhenoMol software is provided for future studies.
BibTeX
@inproceedings{alizadeh2026integration,
  author = {Azar Alizadeh and John Graf and Matthew J. Misner and Andrew Burns and Fiona Ginty and Kevin J. O’Donovan and J. Kenneth Wickiser and N Barringer and Gregory M Freisinger and Neil Hermansen and J.E. Mcdonough and Brian M. Davis and Evelina Loghin and Christine Surrette and Peter Tu and Justin Welch and Oliver Boomhower and Ralf Lenigk and Rachel Sorell and Tyler C. Hammond and Sara Peterson and Alexandre Caron and Leila Safa and Chrystal Chadwick and Stephanie Stacey and J. Jobin and Scott Evans and Rui Xue and Gurvinder Singh Khinda and Eric Williams and Swapnil R. Chhabra and Nhan Huynh and Ernest Fraenkel and Luca Marinelli},
  title = {{Integration of Multiomic and Multi-phenotypic Data Identifies Biological Pathways Associated with Physical Fitness}},
  booktitle = {Communications},
  year = {2026},
  doi = {10.1038/s42003-026-09663-2},
}
Figure