OOPSI: A family of optimal optical spike inference algorithms for inferring neural connectivity from population calcium imaging

Since Ramon y Cajal observed that brains seem to be composed of networks of neurons, neuroscientists and others have wondered about the details of these networks. This work hopefully brings us one step closer towards realizing that goal, at least for small sub-networks. More specifically, this work assumes that the activity of a neural ensemble of around 100--1000 cells has been imaged with calcium indicators; the task is then to infer the most likely set of connections governing these observable neurons. To that end, we built three complementary algorithms. First, a fast, nonnegatively constrained deconvolution filter to infer spike trains online, without requiring any user intervention after image registration and segmentation. Second, a sequential Monte Carlo (SMC) filter can further refine the spike train estimates, and incorporate spike history terms, which are used to infer connectivities. Third, an algorithm to infer the mostly likely connectivity, given the SMC output. While the first two algorithms have been verified using in silico, in vitro, and in vivo experiments, the connectivity inference remains to be confirmed with living cells. That said, the discussion describes several next steps that are currently ongoing.