Sparse Distributed Archetypes Reveal Compressible Network Motifs Underlying Naturalistic Cognition

Area of Science:

• Neuroscience
• Cognitive Science
• Data Analysis

Background:

• Naturalistic cognition arises from complex interactions within distributed brain systems across various scales.
• Characterizing neural organization is difficult due to high-dimensional neural activity obscuring relevant information.

Purpose of the Study:

• To investigate how condition-relevant information is distributed across archetypal representations using Multisubject Archetypal Analysis (MS-AA).
• To explore both spatial and temporal formulations of MS-AA for analyzing naturalistic brain activity.

Main Methods:

• Applied MS-AA to fMRI data from participants listening to intact narratives, word-scrambled audio, and during rest.
• Examined decoding performance and compressibility of information within archetypal representations.
• Conducted network over-representation analyses to identify informative archetypes.

Main Results:

• Decoding performance followed the hierarchy: intact > word-scrambled > rest, indicating preserved structure in archetypes.
• Information was highly compressible, with small subsets of archetypes capturing significant decoding performance.
• Sparse subsets (5-15 archetypes) consistently held disproportionate condition-relevant information and organized subjects into condition-aligned clusters.
• Informative archetypes were distributed mixtures of systems, with default mode and frontoparietal systems being overrepresented.

Conclusions:

• Archetypal motifs distinguishing cognitive states are sparse, distributed subnetworks enriched in higher-order association systems.
• MS-AA offers a valuable framework for studying the compressibility, geometry, and multiscale organization of cognitive brain states.