As new recording methods emerge in neuroscience, new statistical techniques are needed to properly relate neural activity to behavior, a given stimulus, or an internal process. Calcium imaging techniques are now widely used in the ?eld of neuroscience due to their ability to easily record many neurons simultaneously. However, many existing statistical techniques and are formulated for spiketime data. In particular, generalized linear mod- els (GLMs), are developed under the assumption that neural data is in the form of individual spike times. Here, in Aim 1, we propose a framework for extending GLMs to rapidly characterize high-dimensional neural recep- tive ?elds directly from calcium traces. We additionally extend the capabilities of these methods by introducing smoothing priors over receptive ?eld statistics and methods to learn the properties of individual calcium traces on a per-neuron basis that will allow for rapid and ?exible implementation.
In Aim 2, we will use these algorithms on two existing datasets from Ilana Witten (Princeton) and Spencer Smith (UNC) to characterize neural response properties in a spatial reward task and a visual psychophysical task. We will formally quantify both spatial tuning in prefrontal neurons and receptive ?eld properties in a range of cortical visual areas in rodents.
In Aim 3, we will extend our model to elucidate neural properties as functions of internal network dynamics. This will include identifying pairwise neural in?uences by including post-spike ?lters from simultaneously recorded neural units, and additionally learning a latent dynamic signal that may be shared across all neurons in a given cortical re- gion. Due to the many simultaneously recorded calcium traces in our dataset, we will be able to use to our GLM framework to offer insight into how neural networks process and communicate information to different cortical regions for a range of experimental tasks. Our formalized GLM framework will be ef?cient, ?exible, and generalizable across experiments, brain regions, and even animals. We will make our models available for public use upon their publication such that the calcium imaging community can easily adopt them to help answer their own experimental questions.
Our research is central to the mission of the BRAIN initiative as it constitutes a necessary extension of existing statistical tools to new calcium imaging recording techniques. The suite of models and analyses developed here demonstrate a ?exible, ef?cient method for characterizing high-dimensional neural activity in experimental set- tings and understanding neural circuits. Describing the fundamental coding properties of these neural circuits marks a crucial ?rst step in understanding mammalian brain function and ?nding ways to treat and cure mental disease.