Abstract
Animals are not simple input-output machines. Their responses to even very similar stimuli are variable. A key, long-standing question in neuroscience is understanding the neural correlates of such behavioral variability. To reveal these correlates, behavior and neural population must be related to one another on single trials. Such analysis is challenging due to the dynamical nature of brain function (e.g. decision making), neuronal heterogeneity and signal to noise difficulties. By analyzing population recordings from mouse frontal cortex in perceptual decision-making tasks, we show that an analysis approach tailored to the coarse grain features of the dynamics was able to reveal previously unrecognized structure in the organization of population activity. This structure was similar on error and correct trials, suggesting what may be the underlying circuit mechanisms, was able to predict multiple aspects of behavioral variability and revealed long time-scale modulation of population activity.