UNIVERSAL SIGNATURES OF PREDICTION ERROR IN THE <I DATA-PATH-TO-NODE="10,0,0" DATA-INDEX-IN-NODE="48">DROSOPHILA</I> BRAIN: SYNERGY AND REDUNDANCY ACROSS NEURAL HIERARCHIES

Predictive processing has become a leading framework in systems neuroscience, proposing that the brain continuously generates expectations about sensory input and updates internal models through prediction error signals when stimuli deviate from those expectations. While extensively studied in humans and other mammals, it remains unclear whether and how this computational principle operates in simpler, phylogenetically distant nervous systems. In this study, we investigated the distribution of prediction error signals across the brain of Drosophila melanogaster. Using simultaneous electrophysiological recordings from 16 brain regions, we presented flies with expected and unexpected visual stimuli. We then quantified the shared and synergistic structure of prediction error signals using Co-Information, an information-theoretic measure that captures redundancy (shared information) and synergy (information available only in joint activity across regions). Our results revealed patterns of synergistic and redundant prediction error signals in Drosophila that strikingly resembled those observed in common marmosets. In both cases, synergistic and redundant interactions were present within as well as across different stages of the processing hierarchy. Synergy, however, was mainly present between early and late time-points, with respect to stimuli presentation, whereas redundancy was stronger for more proximal time-points. This temporal organization suggests a fundamental property of predictive information processing, where synergistic integration may link early sensory signals with later inferential or feedback processes, and redundancy may reflect robustness to noise. These findings point to conserved principles of predictive computation across brains of vastly different size and structure.