Dynamic Reorganization of Brain Network Information during Emotion and Identity Recognition

Facial expression recognition jointly encodes identity and emotion within large scale brain networks, yet how identity and emotion information are represented and dynamically propagated remains unknown. To address this, we employed a dual-task paradigm integrating emotion and identity recognition with dynamic facial expressions, and recorded MEG signals to track millisecond-level neural dynamics. Using a network-based decoding framework, we identified discriminative spatiotemporal patterns supporting emotion and identity processing. We further applied Partial Information Decomposition (PID) to reveal how redundancy and synergy jointly govern information flow across large scale brain networks. The decoding results revealed distinct spatiotemporal organizations for emotion and identity processing. Emotion decoding exhibited a two-stage temporal processing pattern, with peaks at 180 to 360 ms and 580 to 660 ms, engaging early sensory and later frontoparietal, sensorimotor, and occipitotemporal networks, whereas identity decoding showed a single peak within a stable, bilaterally distributed network spanning temporal, sensorimotor, and frontoparietal regions. Information-theoretic analyses uncovered that rapid processing of facial expressions was dominated by redundant information, whereas integrated information emerged progressively within frontoparietal networks. Together, these findings provide a comprehensive computational account of how the brain dynamically encodes and coordinates emotional and identity information, revealing core principles of facial expression perception.