Characterizing sleep stages through the complexity--weighted-entropy plane in human intracranial data and in a whole-brain model
summary
The gist
Characterizing sleep stages through the complexity--weighted-entropy plane in human intracranial data and in a whole-brain model reveals that each sleep stage occupies a distinct region in this…
In short
The study uses information-theoretic measures, specifically Weighted Permutation Entropy and Statistical Complexity Measure, to map human brain activity onto a complexity-entropy plane. This approach successfully characterizes distinct sleep stages—Wake, N2, N3, and REM—by showing each stage occupies a unique region in this plane. The findings validate the method's robustness across different analytical levels.
Key concepts
- Weighted Permutation Entropy (WPE)
- WPE is an information-theoretic tool used to quantify the complexity of brain signals. It measures how complex and irregular the patterns in neural data are by analyzing permutations of time series. Higher WPE values generally indicate greater complexity, which helps define different brain states.
- Statistical Complexity Measure (SCM)
- SCM is another information-theoretic metric that assesses the statistical properties of neural signals. It helps distinguish between different dynamic regimes in the brain by quantifying how predictable or complex the underlying system dynamics are during sleep stages.
- Complexity-Entropy Plane
- This plane is a conceptual map where different sleep stages are positioned based on their unique complexity and entropy values derived from WPE and SCM. By plotting these measures, researchers can visually and quantitatively identify the distinct dynamical signatures associated with Wake, N2, N3, and REM sleep.
- Whole-Brain Model (AdEx Neurons)
- A simulation model built using Adaptive Exponential Integrate-and-Fire (AdEx) neurons is used to mimic whole-brain dynamics. This model allows researchers to test how the system explores different configurations within the complexity plane, providing a mechanistic understanding of the observed sleep state transitions.
Terminology used across episodes
This episode discusses
- Characterizing sleep stages through the complexity--weighted-entropy plane in human intracranial data and in a whole-brain model · Paper Radio
- Multiscale dynamical characterization of cortical brain states: from synchrony to asynchrony
- State-dependent brain responsiveness, from local circuits to the whole brain
The paper
Characterizing sleep stages through the complexity--weighted-entropy plane in human intracranial data and in a whole-brain model · Read on arXiv
Instituto de Física, Universidade Federal de Alagoas · Instituto de Física Interdisciplinar y Sistemas Complejos, Universitat de les Illes Balears · Institute of Biomedical Research August Pi Sunyer (IDIBAPS) · Paris-Saclay University, CNRS, Paris-Saclay Institute of Neuroscience (NeuroPSI) · ICREA
Approaches based on symbolic information, such as entropy and complexity, have shown that sleep reorganizes the temporal structure of brain activity. However, most of this evidence comes from scalp EEG and group averages. In this work, we characterized sleep stages in intracranial recordings from 65 patients during wakefulness and the REM, N2, and N3 stages, using weighted permutation entropy (WPE) and the statistical complexity measure (SCM). In the complexity--entropy plane, the states formed a clear trajectory from wakefulness to REM, N2, and N3, with decreasing entropy and increasing complexity as sleep deepened. This organization was observed individually in most patients and was recovered more consistently by WPE than by conventional permutation entropy. The trajectory, however, was not the same throughout the cortex. All regions shifted in the same direction, but the occipital cortex followed a much shorter path than the parietal cortex. This result reflects the spatial heterogeneity of the brain during sleep and shows that cortical regions differ not only in the position they occupy in the plane, but also in the path they follow between states. Finally, a whole-brain model based on the human connectome showed that increasing neuronal adaptation shifts the simulated dynamics in the same direction observed in the empirical data, pointing to adaptation as one of the candidate mechanisms involved in this reorganization.
Transcript
Introduction to the show: ident: Genomics Radio. Generated commentary on the latest computational biology and genomics papers.
Ines: I'm Ines.
Marcus: And I'm Marcus, here with Yuki.
Ines: We're looking at the paper: "Characterizing sleep stages through the complexity--weighted-entropy plane in human intracranial data and in a whole-brain model".
Marcus: Here's the gist of the paper. Characterizing sleep stages through the complexity--weighted-entropy plane in human intracranial data and in a whole-brain model reveals that each sleep stage occupies a distinct region in this complexity-entropy plane, demonstrating the robustness of the proposed approach across different levels of analysis The study utilizes information-theoretic quantifiers to characterize sleep stages by mapping them onto the complexity-entropy plane Characterization Methodology
Title and authors: Ines: So we're looking at this paper called "Characterizing sleep stages through the complexity--weighted-entropy plane in human intracranial data and in a whole-brain model." It sounds like they are trying to use information theory to figure out what kind of sleep you're in just by looking at brain activity.
Marcus: Right. The authors are Helena Bordini de Lucas, Leonardo Dalla Porta, Alain Destexhe, Maria V. Sanchez-Vives, Osvaldo A. Rosso, Cláudio R. Mirasso and Fernanda Selingardi Matias > They're taking human brain data from intracranial recordings and whole-brain models to map sleep stages onto a complexity-entropy plane > It’s about moving away from just looking at simple rhythms to understanding the overall information processing dynamics of the brain during sleep, which is pretty deep stuff.
Yuki: From a population genetics angle, this is interesting because if we can distinguish these states robustly in individual data, it suggests that these underlying dynamical regimes are biologically meaningful across different individuals > It’s not just about labeling what they see; it’s about seeing how the whole system behaves differently in each state.
Ines: Exactly. The paper summarizes their main goal as developing a novel tool for sleep stage identification using complexity and entropy measures > They're mapping the brain dynamics onto this plane so that each sleep stage gets its own distinct region > It moves beyond just looking at spectral features like theta or delta waves to capturing the broader structure of the activity.
Marcus: And they use these information-theoretic quantifiers to characterize what's happening in those regions, which is where things get powerful > They are using weighted permutation entropy and statistical complexity measure as input features for an AI classifier, and they report a ninety-two point twenty-five percent accuracy on classifying the sleep stages > That’s a pretty high number for distinguishing between Wake, NREM sleep, and REM sleep based on these metrics.
Yuki: So what does this mean for the wider context? If this complexity-entropy plane truly captures the inherent dynamics of each state, it supports a more dynamic view of sleep rather than just a static categorization > It hints that the underlying mechanisms are distinct in ways that simple frequency analysis might miss.
Ines: That's what I find compelling about it. They show how Wake, NREM sleep, and REM sleep occupy separate areas on this plane, which gives us a quantitative understanding of the inherent dynamics associated with each state > It’s showing us the structural differences in brain information processing during these phases.
Title and authors: Marcus: Then they build a whole-brain model using Adaptive Exponential Integrate-and-Fire neurons to simulate how this complexity plays out across the entire brain > This simulation lets them reproduce and extend those dynamical regimes they saw in the intracranial data, offering a mechanistic account of how the system explores different configurations within that complexity entropy plane.
Yuki: It’s good they are linking it back to a whole-brain model. That takes it from just looking at local activity to thinking about the large-scale organization of brain function during sleep > It connects the microscopic measurements to the macroscopic patterns in a way that feels more biologically plausible.
Ines: They also identify characteristic timescales for state separation by analyzing weighted permutation entropy and statistical complexity measure as a function of embedding delay tau > This helps them pinpoint exactly how fast these states transition between each other, finding that the sensitivity is most pronounced at a time delay of tau equals ten which is fifty milliseconds <ref:2511.09243#pg1>.
Marcus: Fifty milliseconds. That's a concrete temporal scale for when you see the biggest difference in brain dynamics between states > It’s not just saying "it changes over time"; it’s saying "it changes like this fast." They're using that timescale to understand the information processing capabilities and inherent dynamics associated with each state more precisely.
Yuki: For someone listening who isn't a neuroscientist, this means we can think about sleep not as a simple on or off switch between stages, but as a continuous spectrum of different levels of organizational complexity in the brain > It gives us a new way to visualize what’s happening internally.
Ines: And when we put it all together, the overall accuracy they achieved with the support vector machine classifier using these extracted quantifiers is ninety-two point twenty-five percent > That demonstrates that these information-theoretic quantifiers have real potential to discriminate among brain states across the sleep-wake cycle.
Marcus: It shows that this approach isn't just theoretically elegant; it works well when applied to the actual data from intracranial recordings > We’re seeing how these measures can be used as input features for a machine learning model to make a very accurate prediction.
Yuki: So, what does this suggest for how we study sleep in populations? It suggests that if you have large enough datasets, you might not just see average brain activity; you might start to resolve the specific dynamical structures that define different sleep states across individuals > It supports the idea of variability being structured in a complex way.
Ines: The authors acknowledge some limitations, though > They point out that while they can classify these states accurately using their current setup, they are still developing this tool, and it's not yet fully integrated into standard clinical practice for sleep staging.
Title and authors: Marcus: That makes sense. The paper focuses on the classification accuracy of the quantifiers themselves; it doesn't claim it solves every practical problem in a hospital setting right away > It’s a powerful analytical tool, but implementation takes work.
Yuki: I just wonder what comes next for this kind of approach. Does this complexity-entropy plane framework allow us to connect these sleep dynamics to broader cognitive functions or even disease states more directly?
Ines: That's exactly the direction they’re heading > They are using these measures to characterize brain dynamics during different cortical states, which can reveal valuable information about patterns across various cognitive processes, especially memory consolidation and information processing > It opens up avenues for linking sleep structure to those higher-level functions.
Marcus: So we've seen how the authors use the complexity--weighted-entropy plane in human intracranial data and in a whole-brain model to characterize sleep stages, moving from spectral measures to information theory based quantifiers, with a classifier achieving ninety-two point twenty-five percent accuracy > It’s a solid demonstration of using these tools for state identification.
Yuki: It gives us a richer way to think about the inherent dynamics of Wake, NREM sleep, and REM sleep than we had before > I hope future work can leverage this framework to see how these patterns vary across different genetic backgrounds or in relation to specific disorders.
Ines: And that’s where the next step lies > The whole point is that each sleep stage occupies a distinct region on this complexity-entropy plane, demonstrating the robustness of the approach across different levels of analysis > It validates using these information-theoretic quantifiers for characterization.
Marcus: So to wrap up, we see a method that uses weighted permutation entropy and statistical complexity measure to map brain dynamics onto a plane, allowing for high accuracy classification of sleep stages > It’s a tool that helps us understand the inherent dynamics associated with each state better than before.
Yuki: It’s about seeing the underlying structure of brain activity in sleep states as an information processing phenomenon, not just as a set of simple waves or rhythms > That's a big shift for how we conceptualize sleep biologically.
Ines: Exactly. We are looking at the full picture now, from the local measurements up to the whole-brain simulation and classification performance of ninety-two point twenty-five percent > This paper is showing us that information theory has a lot to offer when analyzing complex biological systems like the human brain.
The paper's summary: Ines: So, this paper is about using information theory—specifically complexity and entropy measures—to map human brain data onto a plane, showing that each sleep stage has its own spot on that plane.
Marcus: Right. They're not just looking at simple waves anymore; they’re mapping the whole dynamic behavior of the brain during different sleep states onto this complexity-entropy plane to see how those stages actually sit relative to each other.
Ines: The core idea is that Wake, NREM sleep, and REM sleep occupy distinct regions on this plane, which gives us a quantitative way to understand the inherent information processing capabilities of each state.
Marcus: It's about moving past just labeling the waves and getting a real picture of how the brain's dynamics change structurally when you switch between those stages.
Yuki: From my perspective, this is interesting because if we can see these distinct regions on a plane, it suggests that these different sleep states have fundamentally different underlying organizational rules for information processing.
Ines: Exactly. The study uses these measures to classify the cortical states with high accuracy—about ninety-two point twenty-five percent when using a support vector machine classifier on the extracted features.
Marcus: That’s a strong result, especially when you consider that they are using these information theory quantifiers as inputs for the classifier; it really shows how much discriminating power these measures have for distinguishing brain states across the sleep-wake cycle.
Yuki: What this means is that if we can reliably separate Wake from NREM or REM based on these dynamics, it supports a deeper view of sleep as a spectrum of different types of brain organization rather than just distinct rhythms.
Ines: It’s about showing that the complexity and entropy measures are robust enough to work across different levels of analysis, from just looking at local intracranial data up to testing them in a whole-brain model.
Marcus: They build this whole-brain model using adaptive exponential integrate-and-fire neurons, and it lets them simulate those specific dynamical regimes they saw in the actual intracranial recordings.
Yuki: That simulation part is key because it connects the microscopic measurements to a larger system, allowing them to mechanistically explain how the brain explores those different configurations within that complexity plane.
Ines: They also pinpoint specific timescales for how states separate by looking at how sensitive the weighted permutation entropy and statistical complexity measure are as you change the embedding delay tau.
Marcus: They found that this sensitivity peaks at a time delay of ten, which is fifty milliseconds, which gives them a concrete temporal scale for when these state differences become most pronounced.
Yuki: So we can quantify not just *if* a brain is in REM sleep, but how its internal dynamics are structured in terms of complexity and entropy during that period.
Ines: And what this changes for us is that it provides a new framework to think about sleep not as a simple on or off switch between stages, but as a continuous spectrum of different organizational complexities in the brain.
Marcus: It’s about seeing the underlying structure of brain activity in sleep states as an information processing phenomenon, which is a significant shift from just looking at simple frequencies.
Yuki: This opens up avenues for linking these sleep structures to broader cognitive functions and potentially to disease states, because if we understand the dynamics of each state so well, we can start asking more complex questions about what those dynamics actually *do*.
The paper's improvements: Tom: So, we're talking about how they suggest improving this whole approach to characterizing sleep stages using complexity and entropy measures on human brain data.
Ines: The improvements focus on making the AI system even better at automatically classifying cortical states by using those information theory quantifiers like weighted permutation entropy and statistical complexity measure.
Marcus: That means the system can use those metrics as inputs for a support vector machine, which they say gives a very high accuracy, about ninety-two point twenty-five percent when it’s distinguishing between different cortical states.
Yuki: From a population genetics angle, that high accuracy suggests that these underlying dynamical signatures are strong enough to be reliable features for individual classification.
Ines: They also suggest improving the whole-brain model by using adaptive exponential integrate-and-fire neurons, which lets the system reproduce and extend the range of brain dynamics observed in the intracranial data.
Marcus: That simulation capability is important because it moves beyond just analyzing local brain activity to actually modeling how those dynamics scale up across an entire brain structure.
Yuki: It’s about making sure that when we look at a specific dynamic regime, we have a biological mechanism behind it in the model, not just a mathematical fit.
Ines: Another key improvement is identifying characteristic timescales for state separation by analyzing how the weighted permutation entropy and statistical complexity measure change as you vary the embedding delay tau.
Marcus: They pinpointed that this sensitivity is most pronounced at a time delay of ten, which translates to fifty milliseconds, giving them a specific temporal scale for when these states are most distinctly separated.
Yuki: That helps ground the findings in actual brain timing; it’s not just "it changes over time," it’s "it changes like this fast."
Ines: So, essentially, they're improving the method by building a more robust computational tool that can accurately classify states and provide a mechanistic simulation of those states.
Marcus: The implication here is that we have a much stronger set of tools for analyzing sleep data than before; it’s about using these information theory measures to get high-fidelity predictions and structural models.
Yuki: This gives us a more detailed view of the internal organization during sleep, which could eventually link these patterns to broader cognitive functions or even how brain structure might change in different conditions.
Conclusion: Tom: So, to wrap up, this paper on "Characterizing sleep stages through the complexity--weighted-entropy plane in human intracranial data and in a whole-brain model" shows how information theory metrics can map complex brain dynamics onto a plane where each sleep stage has its own distinct region.
Ines: It’s about using weighted permutation entropy and statistical complexity measure to classify states with ninety-two point twenty-five percent accuracy, which is pretty strong for distinguishing between Wake, NREM sleep, and REM sleep.
Marcus: And the whole point is that this method works on two levels—using those quantifiers for classification, and then building a whole-brain model to simulate those specific brain dynamics.
Yuki: This framework helps us see sleep not just as a set of rhythms, but as a continuous spectrum of organizational complexity within the brain.
Ines: Exactly; it’s about getting a quantitative understanding of the inherent information processing capabilities and dynamics associated with each state, moving past simple frequency analysis.
Marcus: And they found that analyzing how sensitive those measures are to embedding delay tau tells us exactly when these states separate, which is around fifty milliseconds.
Yuki: That temporal scale gives us a concrete way to think about the transition between sleep stages in terms of brain organization.
Ines: So, the implication is that we have a powerful new tool for analyzing sleep data by treating brain activity as an information processing system rather than just looking at simple wave patterns.
Marcus: It’s a solid demonstration of how these information theory quantifiers can discriminate among different brain states across the sleep-wake cycle using machine learning.
Yuki: It supports the idea that if we have large enough datasets, we might be able to resolve these specific dynamical structures across individuals.
Ines: So, while it’s a powerful analytical tool for classification and modeling, they do flag that it’s still developing this tool and isn't fully integrated into standard clinical practice yet.
Marcus: Right. It's a sophisticated analysis method, but implementation still takes work before we see it all the way into a clinic.
Yuki: This paper sets up the foundation for linking these sleep structures to broader cognitive functions down the road, which is where I’m most interested in seeing this research go next.
More episodes
- 2610.33881-Sensorimotor behavior from an untrained brain-and-nerve-cord connectome
- 2610.40344-Morphological Features of Murine Pulmonary Arterial Networks for Diagnosing Vascular Disease
- Daily Summary for 2026-11-19
- 2609.10193-Are You Learning Biological Signal or Shortcuts? Auditing and Mitigating Bias in Protein-Protein Interaction Datasets
- 2610.38236-Generative discovery of chemical reaction networks
- 2507.11176-An Interpretable AI framework Quantifying Traditional Chinese Medicine Principles Towards Enhancing and Integrating with Modern Biomedicine
- The day in conversation for 2026-11-19
- Daily Summary for 2026-11-19
- 2504.03590-Towards a Unified Framework for Determining Conformational Ensembles of Disordered Proteins
- 2603.15217-A multiscale discrete-to-continuum framework for structured population models