Comments on “The Human Organism as an Integrated Interaction Network: Recent Conceptual and Methodological Challenges”
by Klaus Lehnertz, Timo Bröhl and Thorsten Rings
The field of Network Physiology is built on a compelling and intuitive premise: to understand the human organism not as a collection of isolated organs, but as an integrated, evolving network of dynamical interactions1111. This framework, which treats organs as vertices and their communications as edges, promises to revolutionize our understanding of health and disease by moving beyond reductionist approaches.
However, a critical 2020 perspective by Lehnertz, Bröhl, and Rings in Frontiers in Physiology serves as a vital “state of the field” and a sober reminder of the profound methodological and conceptual challenges that lie ahead. The authors meticulously deconstruct the path from raw physiological recordings to a meaningful network representation, highlighting the many assumptions and unsolved problems we face. This is not a dismissal of the field, but rather a necessary and constructive critique calling for rigor and innovation.
As Lehnertz et al. frame it, the entire endeavor is a fundamental “inverse problem” : we are attempting to infer the properties of a complex, underlying system (the network) from limited, often noisy, observations (time-series data). The authors argue that this inference is far from straightforward. We explore their three main categories of challenges.
Challenge 1: Can We Accurately Characterize Interactions?
The first hurdle is quantifying the interaction—the edge—between two physiological systems (e.g., brain and heart) from their recorded time series. While a vast toolkit of bivariate and multivariate time-series-analysis techniques exists—drawn from statistics, information theory, and non-linear dynamics —they all run into a fundamental mismatch.
The authors astutely point out that most analysis techniques assume stationarity, meaning the system’s properties are stable over the observation time. Yet, by its very nature, physiology is inherently non-stationary, and interactions are often transient. A coupling that exists for 10 seconds may be missed entirely or averaged into noise by a method that assumes stability over 10 minutes.
Furthermore, these systems operate on vastly different timescales (e.g., milliseconds for neural activity, hours for hormonal changes) , and we often lack knowledge of the true, time-varying delays between them. This complexity is compounded by the lack of reliable surrogate testing for many methods, especially those assessing causality or coupling functions, making it difficult to know if an observed interaction is statistically significant or an artifact.
Challenge 2: How Do We Even Build the Network?
Even if we had perfect interaction measures, Lehnertz et al. identify critical, unresolved issues in the very definition of the network’s components.
On Vertices (Nodes)
What, precisely, is a “vertex” in the human network? We intuitively say “the brain” or “the heart,” but in practice, a vertex is a sensor—an EEG electrode, an ECG lead, a blood pressure monitor. This “sensor-space” representation is an assumption, not a given. Furthermore, the observables are of completely different physical and chemical natures (voltage, pH, temperature, pressure), each capturing dynamics on its own characteristic timescale. The article rightly asks: how do we create an integrated network from such heterogeneous data?
On Edges (Connections)
This is perhaps the most significant challenge. The interactions we measure are functional (inferred from data), not necessarily structural (a physical nerve tract or blood vessel). The authors highlight two massive, related problems:
- Direct vs. Indirect Interactions: Is the brain directly signaling the heart, or is their apparent coupling mediated by a third, unobserved system (e.g., the lungs or a hormonal response)?. Most techniques struggle to reliably distinguish this, a problem known as transitivity.
- Common Source Contamination: Spurious interactions can appear simply because two sensors are recording a common, underlying source (e.g., volume conduction in EEG).
On Network Type
Once vertices and edges are defined, what kind of network do we build? A simple binary (connected/unconnected) network requires a threshold, but there is no accepted criterion for choosing one. A weighted network is more informative, but how do we normalize weights from different measures?.
The most difficult problem is creating a weighted and directed network. Lehnertz et al. stress that interaction strength and direction are distinct properties. The modulus of a directionality index (like Granger causality) cannot simply be interpreted as strength, yet this is a common pitfall. How to properly merge these two properties into a single, meaningful network remains an open question.
Challenge 3: Are We Analyzing with the Right Tools?
Finally, the authors caution against the uncritical application of standard graph-theory measures to these physiological networks. Characteristics like the clustering coefficient and mean shortest path—famously used to define “small-world” networks—were developed for simple, binary, undirected graphs. Their extension to weighted and directed networks is “usually not straightforward” , and different definitions can yield different results. This ambiguity has led to considerable debate about whether findings like “small-worldness” in brain networks are robust topological features or artifacts of analytic choices.
Conclusion: A Call for Concerted Effort
The work by Lehnertz et al. is not a pessimistic conclusion. It is a call to action. It clearly articulates the need for a new generation of tools and concepts:
- Sensing technologies capable of time-locked, multimodal recordings.
- Analysis techniques that can handle non-stationary, transient, and multiscale data.
- Network concepts better suited to this heterogeneity, such as multilayer or multiplex networks.
The authors leave us with the single most important question that everyone in this field should consider: ultimately, will the network framework “tell us anything new about the human organism we did not know before?”. Answering that question requires us to first honestly confront and overcome the challenges they have so clearly defined.
Link to the article https://pubmed.ncbi.nlm.nih.gov/33408639/