HyperNESS: Decomposing Inter- and Intra-Brain Neural Contributions in Dual-Brain Hyperscanning
Instructors: Chuyang Sun
About this course
Hyperscanning research has advanced our understanding of social interaction, yet a fundamental challenge remains: traditional inter-brain synchronization measures cannot distinguish genuine social coordination from parallel individual responses to shared stimuli. This conflation prevents mechanistic conclusions about how social interaction shapes individual neural processes.
This minicourse introduces HyperNESS (Hyperscanning Network Estimation via Source Separation), a novel framework extending FREQ-NESS (Rosso et al., 2025, Advanced Science) to dual-brain hyperscanning. By treating simultaneously recorded data from two participants as a unified 44-dimensional space and applying Generalized Eigenvalue Decomposition (GED) to the joint covariance matrix, HyperNESS enables exact mathematical decomposition of neural variance into intra-brain (individual processing) and inter-brain (social coordination) contributions — a distinction conventional methods cannot achieve.
We demonstrate HyperNESS using fNIRS hyperscanning data from 34 dyads during cooperative gaming, where inter-brain and intra-brain components show frequency-specific profiles and functionally distinct relationships with reward processing. The minicourse covers mathematical foundations, step-by-step MATLAB implementation, and statistical inference via cluster-based permutation testing. Participants will leave equipped to apply HyperNESS in their own hyperscanning research.
What will you learn?
- Understand why conventional inter-brain synchronization measures conflate social coordination and individual task processing, and why this distinction matters for social neuroscience research
- Grasp the mathematical logic of Generalized Eigenvalue Decomposition (GED) as applied to frequency-resolved brain network estimation
- Implement the HyperNESS pipeline in MATLAB — from dual-brain data integration through GED computation, variance decomposition, and ROI-based spatial analysis
- Quantitatively separate inter-brain and intra-brain neural contributions from hyperscanning data and interpret what each component means functionally
- Apply cluster-based permutation testing and FDR correction for statistically robust inference across frequency spectra
Course details
Course duration
1 hours
Prerequisites and/or other notes
N/A
Delivery plan
0-10 min: Introduction
10–20 min: Theoretical Foundations
20–40 min: The HyperNESS Framework
40–45 min: Implementation Walkthrough
45–55 min: Empirical Application
55–60 min: Q&A
Why should you enroll in this course?
Hyperscanning — the simultaneous neuroimaging of two or more interacting individuals — has become a central tool in social neuroscience for studying the neural basis of real-time human interaction. A growing body of work has documented inter-brain synchronization across diverse social contexts, from cooperative tasks and joint attention to communication and emotional exchange (Czeszumski et al., 2020; Kingsbury & Hong, 2020). Yet despite this progress, the field faces a persistent and underappreciated methodological challenge: conventional inter-brain synchronization measures cannot distinguish whether correlated neural activity across two brains reflects genuine social coordination — arising from mutual interpersonal influence — or simply parallel individual responses to the same shared stimuli (Hamilton, 2021). This conflation of inter-brain and intra-brain contributions means that increased synchrony in a dyadic task cannot, by itself, tell us whether social interaction is actually doing any neural work.
This limitation is not merely technical — it has direct consequences for theory. Questions about how social environments shape individual neural processes, such as how interacting with others alters reward processing, cognitive control, or risk-taking, require precisely the kind of clean separation that current methods do not provide. Without it, social neuroscience risks drawing mechanistic conclusions from measures that are inherently ambiguous in their neural origins.
HyperNESS addresses this gap by extending the FREQ-NESS framework (Rosso et al., 2025, Advanced Science) — which uses Generalized Eigenvalue Decomposition to isolate frequency-specific brain networks — to dual-brain hyperscanning scenarios. By treating simultaneously recorded data from two participants as a single unified high-dimensional system rather than two independent ones, HyperNESS preserves the inter-brain covariance structure that conventional individual-first approaches discard. This enables an exact mathematical decomposition of total neural variance into intra-brain contributions from each participant and inter-brain contributions arising from their interaction, within a single coherent analytical framework.
The goals of this minicourse are threefold. First, participants will develop a solid conceptual understanding of why the inter-brain / intra-brain separation problem matters and how existing hyperscanning methods fall short. Second, they will learn the mathematical foundations of HyperNESS — including the GED procedure, the block covariance structure of dual-brain data, and the variance decomposition logic — in a way that is accessible without sacrificing rigor. Third, they will gain practical experience implementing the full HyperNESS pipeline in MATLAB, including frequency-resolved network estimation, variance decomposition, cluster-based permutation testing, and neural-behavioral correlation analysis. By the end of the course, participants will be equipped to apply HyperNESS to their own hyperscanning datasets, whether collected with fNIRS, EEG, or MEG.