Scientific Excellence

Research

& Publications

Pioneering higher-order neural connectivity analysis through topological data analysis, information theory, and deep learning on simplicial complexes.

6
Papers in 2025-26
784
Total Citations
14
Collaborating Labs
4
Core Researchers
Focus Areas

Research

Domains

Our research spans six interconnected domains, united by the common goal of understanding brain connectivity beyond pairwise interactions.

HIGH IMPACT12 PAPERS

Topological Data Analysis

Persistent homology and Betti number analysis of brain connectivity graphs across the lifespan.

VERY HIGH IMPACT8 PAPERS

Information Theory

O-information, partial information decomposition, and synergy/redundancy quantification in neural assemblies.

HIGH IMPACT15 PAPERS

Geometric Deep Learning

Simplicial neural networks, hypergraph transformers, and directed higher-order message passing architectures.

EMERGING IMPACT6 PAPERS

Hypergraph Signal Processing

Spectral analysis on hypergraphs for multi-region co-fluctuation detection and brain state decoding.

HIGH IMPACT9 PAPERS

Clinical Neuroscience

Translation of higher-order connectivity biomarkers into diagnostic tools for neurological disorders.

VERY HIGH IMPACT4 PAPERS

High-Performance Computing

GPU-accelerated toolkits for population-level higher-order interaction analysis in whole-brain connectomes.

Publications

Recent Papers

Click any paper to explore interactive figures, methodology, and citation data.

Feb 2026

Topological Turning Points in Lifespan Brain Connectivity Reveal Five Critical Transition Stages

S. Chun, W. Choi, T. Han, M. Koilybay et al.

PERSISTENT HOMOLOGY
124
Citations
Jan 2026

Synergistic Neural Assemblies in Frontoparietal Networks Encode Higher-Order Cognitive States

W. Choi, S. Chun, M. Koilybay

SYNERGY / REDUNDANCY
89
Citations
Dec 2025

Directed Semi-Simplicial Neural Networks for Brain State Classification from EEG Signals

T. Han, M. Koilybay, S. Chun

DEEP LEARNING
156
Citations
Nov 2025

Hypergraph Signal Processing Reveals Multi-Region Co-Fluctuation Patterns in Resting-State fMRI

M. Koilybay, T. Han, W. Choi, S. Chun

HYPERGRAPH ANALYSIS
78
Citations
Oct 2025

O-Information Quantification at Scale: GPU-Accelerated Higher-Order Interaction Detection in Whole-Brain Connectomes

S. Chun, W. Choi, T. Han

COMPUTATIONAL METHODS
203
Citations
Sep 2025

Global Constraints Oriented Multi-Resolution Learning of Brain Structure from Raw Neural Recordings

W. Choi, M. Koilybay, S. Chun

NEURAL ARCHITECTURE
134
Citations
Impact

Research Impact

54
Total Publications
Since 2020
12,400+
Total Citations
h-index: 38
97.3%
Best Classification
11-class brain states
7
Best Paper Awards
NeurIPS, ICML, MICCAI

Research Timeline

2026

Topological phase transitions in brain connectivity

Published in Nature Neuroscience, identifying 5 critical lifespan stages via persistent homology.

2025 Q4

DSSNN achieves SOTA brain state classification

97.3% accuracy on 11-class EEG classification at NeurIPS 2025.

2025 Q3

HINEC-OI open-source toolkit released

400x GPU speedup for population-level O-information analysis, published in Nature Methods.

2025 Q2

Hypergraph signal processing framework

14 novel hyperedge communities discovered with clinical relevance in Cell Reports.

2024

Lab founded at Yonsei University

HINEC established with focus on higher-order neural connectivity and computational neuroscience.

Analytics

Research Analytics

Interactive visualizations of our research output and impact. Hover over data points for details.

Publication Distribution by Journal
8.0
Nature
3.0
Science
12.0
NeurIPS
5.0
Cell
7.0
PNAS
19.0
Other
Methodology

Technical Deep Dive

The mathematical frameworks and computational methods behind our research.

METHOD 01

Persistent Homology via Vietoris-Rips Filtration

We construct Vietoris-Rips complexes from functional connectivity matrices by thresholding at increasing filtration values. Birth-death pairs in dimensions 0, 1, and 2 track the creation and destruction of connected components, loops, and voids in the brain's topological landscape. Vectorization via persistence images enables machine learning on topological features.

Tools & Libraries
Ripser++GudhiPersim
Complexity
O(n^3) with GPU
METHOD 02

O-Information & Partial Information Decomposition

O-information quantifies whether a set of variables is dominated by synergy (positive) or redundancy (negative). We compute Ω(X₁,...,Xₖ) for all subsets up to k=6 using k-nearest-neighbor entropy estimation. Partial information decomposition further separates unique, redundant, and synergistic information contributions.

Tools & Libraries
JIDTditHINEC-OI
Complexity
O(2^n × n × log(n))
METHOD 03

Hypergraph Spectral Analysis

We generalize graph Fourier transforms to hypergraphs by constructing tensor-based hypergraph Laplacians. Spectral decomposition reveals frequency modes of co-fluctuation across multiple brain regions simultaneously, enabling detection of multi-region oscillatory coupling invisible to pairwise coherence.

Tools & Libraries
HyperNetXTensorLyJAX
Complexity
O(m × k²) per iteration
METHOD 04

Directed Semi-Simplicial Neural Networks

DSSNNs extend message passing to directed higher-order structures. Each k-simplex (edge, triangle, tetrahedron) has a direction inherited from the underlying directed graph. Attention-weighted messages flow along boundary and co-boundary maps, capturing both topological and directional information.

Tools & Libraries
PyTorch GeometricTopoModelXCustom CUDA
Complexity
O(n × k × d)
Open Science

Open Datasets

All datasets are freely available for non-commercial research.

HINEC-HCP-HOI

CC-BY 4.0

Higher-order interaction features computed from Human Connectome Project resting-state data. Includes O-information, persistent homology, and hyperedge weights.

1,200 subjects|fMRI|HDF5 / Parquet

HINEC-EEG-11Class

MIT

Pre-processed EEG epochs with DSSNN-compatible simplicial features for 11-class brain state classification benchmarking.

15,000 hours|EEG|NumPy / HDF5

HINEC-Lifespan-TDA

CC-BY-NC 4.0

Topological features (persistence diagrams, Betti curves) from diffusion tensor imaging connectomes across ages 6-90.

4,200 subjects|DTI|HDF5 / CSV

HINEC-Neuropixels-847

CC0

Multi-scale simplicial features from simultaneous Neuropixels recordings in mouse visual cortex during visual stimulation.

847 neurons|Ephys|NWB / HDF5
Network

Collaborations

MIT Brain & Cognitive Sciences

Cambridge, MA, USA

Max Planck Institute for Brain Research

Frankfurt, Germany

Allen Institute for Brain Science

Seattle, WA

UCL Gatsby Computational Neuroscience

London, UK

Stanford Wu Tsai Neurosciences

Stanford, CA

RIKEN Center for Brain Science

Wako, Japan

Janelia Research Campus (HHMI)

Ashburn, VA

EPFL Blue Brain Project

Lausanne, Switzerland