Brain wide distribution of prior belief constrains neural models of probabilistic inference

Despite numerous studies, the neural basis of Bayesian inference during decision-making remains unclear. To address the question of where and how prior knowledge about a task is combined with sensory observations, we examine brain-wide neural recordings collected by the International Brain Lab (IBL). In the IBL task, mice indicate the location of a visual grating stimulus (left or right). Crucially, the prior probability that the stimulus appears on the left flips between 20% and 80% in blocks of variable length. The mice leveraged the block structure to improve their decision accuracy, notably performing better than chance (using the prior) when the grating contrast is set to zero. Using IBL’s complete dataset of Neuropixels electrophysiological recordings and wide-field cortical calcium imaging, we decoded the Bayes optimal prior estimate from most of the brain regions in the Allen atlas (N ~ 160 regions). We establish in both electrophysiology and wide-field imaging that the prior is widely represented throughout the mouse brain in a pre-stimulus epoch. In agreement with previous work, we find it present in particular in high level cortical areas such as ACC or OFC. However, it is also seen throughout substantial portions of the rest of the brain, including early sensory cortical and subcortical regions, such as primary visual cortex and lateral geniculate nucleus, as well as motor-related regions such as M1 and the cerebellum. This widespread representation of the prior argues for a neural model of inference involving loops between areas, as opposed to a model in which prior information is incorporated only in decision making areas, thus shedding light on a critical aspect of Bayesian inference in neural circuits. This study offers the first brain-wide perspective on prior encoding, underscoring the importance of using large scale recordings on a single standardized task.