Representational geometry of sounds in the auditory cortex explains biases in perceptual decision-making in mice

Decision-making is often biased, leading to non-optimal outcomes. However, how these biases emerge is only poorly understood. Here, we leverage the mouse auditory system, testing perceptual discrimination of pulsed sound stimuli in a categorical go/no-go task, and observe systematic choice biases to specific sounds shared across individuals. We parametrically characterize the pulsed sounds using an interpretable set of stimulus features and identify those that effectively capture stimulus-evoked activity as measured by large-scale two-photon calcium imaging in the auditory cortex. We find that stimulus features are encoded with varying degrees of representational entanglement. Linking neural and behavioral assessments, we show that the feature set of a given stimulus predicts perceptual decisions and associated biases. Together, our findings highlight representational architectures as relevant underpinnings of perceptual decision-making.