From Calcium Sensitive Fluorescence Movies to Spike Trains

Abstract

Motivation Calcium imaging is quickly becoming a prominent paradigm to collect data in neuroscience. To maximally utilize the power of this technique, complementary analytical tools can be built. Goal We aim to develop analytical tools to facilitate inferring spike trains from fluorescent observations, fitting tuning curves, and inferring population connectivity, given only short sequences of possibly very noisy, low temporal resolution, and saturating fluorescence images. Solution By framing the problem as a state-space problem, we can utilize tools developed by the statistics community for related problems. In particular, we develop a (i) fast filter utilizing a tridiagonal trick and interior point methods to approximate the MAP spike train, (ii) particle filter to infer the probability of spiking in each frame, (iii) a population version of our particle filter to infer connectivity. Conclusions Our fast filter can accurately approximate the most likely spike train given the fluorescence data inO(T ) time. When fluorescence saturates, stimuli are present, or temporal resolution is unsatisfactorily slow, our particle filter can infer the probability of a spike in each time bin. If a small population of neurons are imaged simultaneously, our population particle filter can learn the effective connectivity of the observable neurons. References 1) Vogelstein JT, Watson BO, Packer AM, Yuste R, Jedynak B, and Paninski L. Spike inference from calcium imaging using sequential Monte Carlo methods. In Press at Biophysical Journal. 2) Vogelstein JT, Babadi B, Packer AM, Yuste R, and Paninski L. Fast methods for spike inference from calcium imaging. In preparation. Acknowldgments Support for JTV was provided by NIDCD DC00109. LP is supported by an NSF CAREER award, by an Alfred P. Sloan Research Fellowship, and a McKnight Scholar Award. 1. simple