Breaking Biometric Babel
Making sense of time-series physiological data with PhysioNexus, an open-source tool that transforms complex data into intuitive network visualizations showing cause-and-effect relationships.
Enjoy this piece? Show your support by tapping the ❤️ in the header above. It’s a small gesture that goes a long way in helping me understand what you value and in growing this newsletter. Thanks!
This is a non-technical overview of PhysioNexus, an open-source tool I created for analyzing time-series physiological, environmental, and external load data. Those interested in the technical details, including implementation instructions, network visualization interpretation, and metric calculations, can find comprehensive documentation in the GitHub repository.
The evolution of physiological monitoring has followed a predictable path. We started with single metrics—heart rate monitors strapped to people’s chests—and gradually added more sensors, more data streams, and more complexity. The industry's solution to better physiological insights has consistently been "more data." But after a decade working with Olympians, professional athletes and teams, and human performance groups within the DoD, I've witnessed this approach reach its logical endpoint: a deluge of disconnected metrics that fail to reveal how our bodies actually function as integrated systems.
The fundamental limitation isn't in our measurement capabilities — after all, we can simultaneously track dozens of biometrics with nearly perfect precision. Rather, the limitation is our analytical framework. It's as if we've built one of the most advanced microscopes but are looking through it with one eye closed. For example, when an elite cyclist’s muscle oxygenation plummets during a climb traditional analysis can tell us that this correlates with rising blood lactate and heart rate— but correlation isn't causation. Which variable is driving the cascade? Which is merely responding?
This analytical dead-end led to my developing PhysioNexus, an open-source tool that moves beyond correlation to map causal networks in physiological data. PhysioNexus draws on bioinformatics techniques similar to those used for analyzing gene co-expression and protein-protein interaction networks, but adapts these approaches specifically for time-series physiological data. The approach uses Granger causality, which is a framework that defines causation through predictive power— If knowing the history of variable A helps us predict future values of variable B (beyond what past values of B alone can tell us), then A “Granger-causes” B. This directional relationship carries important information that correlation analysis misses entirely.
Consider a concrete example: In traditional analysis, seeing that respiratory rate rises as muscle oxygenation falls simply establishes correlation (in this case negative correlation). PhyioNexus, however, can determine that in a specific individual muscle oxygenation changes consistently precede and predict respiratory rate changes, suggesting a causal relationship where local tissue-level changes drive systemic cardiorespiratory responses— not the other way around. This distinction fundamentally changes how we understand and act on the data.
What makes physiological systems particularly challenging to analyze is their circular, interconnected nature. The body doesn't operate in neat linear pathways but through complex feedback loops— like one giant recursive algorithm. When muscle oxygenation drops, ventilation increases, which affects blood pH, which influences oxygen binding affinity, which circles back to muscle oxygenation. PhysioNexus navigates this complexity through multivariate testing, identifying when combinations of variables jointly cause changes in others—something beyond the reach of conventional analytical methods.
The visual outputs from PhysioNexus transform abstract statistical relationships into intuitive network diagrams where nodes represent physiological variables and connecting arrows show causal relationships. Blue connections indicate positive relationships (one variable increases, the other follows), while red connections show negative relationships (one increases as another decreases). The thickness of these connections represents the statistical strength of the relationship.
These visualizations aren't just static charts — they’re functional, navigable, tools revealing the operational architecture of an athlete's physiology. Just as individual genomes contain surprises that defy population averages, individual physiomes often contradict textbook models. In one professional cyclist I studied, the network analysis showed something surprising: muscle oxygenation, blood lactate, and muscle recruitment, not heart rate, were the primary causal drivers of the entire physiological cascade during high-intensity exercise. This "peripheral dominance" profile contrasted sharply with the typical "central cardiovascular" regulation pattern seen in most endurance athletes.
Such insights radically transform training approaches. For athletes with peripherally-dominant causal networks, prioritizing interventions targeting local muscle adaptations (capillarization, mitochondrial density, lactate clearance) will yield better results than focusing primarily on central cardiovascular improvements. The causal network essentially provides a physiological blueprint unique to each athlete, showing precisely which systems drive performance and which merely respond.
This approach extends far beyond sports performance. In clinical physiology, PhysioNexus can map how causal relationships transform when comparing healthy versus pathological conditions or when examining adaptation to environmental stressors like altitude or heat. The changes in causal architecture often reveal adaptation mechanisms hidden to conventional analysis.
While the theoretical foundations draw from network biology principles used to analyze genomic and proteomic datasets, PhysioNexus has been specifically optimized for the unique challenges of time-series physiological data, where temporal dynamics and real-time feedback mechanisms create complex causality patterns not typically seen in static biological networks.
The current implementation of PhysioNexus does have its limitations though — like all causal inference methods, it can be misled by unmeasured variables that drive apparent relationships between measured ones. It also focuses primarily on linear relationships. However, future development will extend it’s capabilities into nonlinear causality using transfer entropy, multi-scale analysis to capture relationships operating at different timescales, and dynamic network analysis to examine how causal networks evolve throughout exercise. The ultimate goal is transforming these causal networks into predictive differential equation models—creating virtual physiological testing grounds where interventions can be simulated before implementation.
What excites me most about this approach is its potential to fundamentally change how we think about human physiology. Rather than viewing the body as a collection of independent systems, PhysioNexus reveals the interconnected causal architecture that defines our physiological function. It transforms our understanding from isolated metrics to integrated networks, from correlations to causes.
The implications extend beyond improving athletic performance. This causal network approach could transform how we detect early warning signs of overtraining or illness before traditional markers appear, how we understand individual variability in response to training or treatments, and how we personalize interventions based on unique physiological architectures.
As we continue collecting more physiological data from wearable biosensors, smart clothing, and environmental monitoring devices, our limitation to understanding will not be our measurement capabilities— it will be our analytical frameworks. Moving beyond correlation to causality represents the necessary evolution in how we understand complex physiological systems. PhysioNexus represents one step in that direction, with comprehensive documentation and the source code available on GitHub for researchers and practitioners to implement, test, and extend.
After all, in physiology as in science generally, the most important question isn't just what happens, but why it happens. And answering "why" requires moving beyond correlation to causality. The networks are waiting to be mapped.
Interested in working together? I advise small companies, startups, and VC firms on topics ranging from biosensor development, multiomics and biometric data analysis, network modeling, and product strategy. Contact eva♦peiko♦@gmail.com (replace the ♦ with n) for consulting inquiries or newsletter sponsorship opportunities.
Wonderful article! It reminds of Judea Pearl's the Book of Why?