Trial matching: capturing variability with data-constrained spiking neural networks

Simultaneous behavioral and electrophysiological recordings call for new methods to reveal the interactions between neural activity and behavior. One main challenge is to create an interpretable model of the co-variability of spiking activity and behavior across trials. In this talk, we will present our model of a mouse cortical sensory-motor pathway in a tactile detection task reported by licking. The core innovation of our work was to fit a large recurrent spiking neural network (RSNN) with the same trial-to-trial variability as in the data by harnessing the intrinsic noise of the model. Our solution relied on optimal transport to define a distance between the distributions of generated and recorded trials. This technique was applied to artificial data and neural recordings covering six cortical areas. We found that the resulting RSNN can generate realistic cortical activity and predict jaw movements across the main modes of trial-to-trial variability. Our analysis also identified an unexpected mode of variability in the data corresponding to task-irrelevant movements of the mouse. We hope that such data-driven approaches will help us to get one step closer to a mechanistic understanding of such sensor-to-motor transformations.