Advanced computational analysis of calcium imaging data

Abstract

Calcium imaging has emerged as a promising technique to indirectly measure the neural activity of populations of neurons, thereby enabling the functional analysis of these assemblies. This technique has been used to study the neural mechanisms and phenomena under a wide range of experimental and behavioral conditions. These include, as a point of interest for this thesis, the sparsification phenomenon occurring during early neocortical development, where population activity undergoes a dramatic transition from largely synchronized (cluster activity) to a relatively sparse mode around the time of eye-opening in rodents. Despite its wide-range of applications, the fluorescence traces (time-series) recorded by this technique are temporally smeared. This restricts the accurate reconstruction of quantities of interest such as spike times. Therefore, several spiking activity reconstruction methods have been introduced, where most of them are limited in dealing with issues such as variability in the kinetics of fluorescence transients, fast processing of long-term data, measurement noise and high firing rates. In addition, none of these methods provide insights into the underlying, intrinsic neuronal dynamics or biophysical parameters. As the first goal of this thesis, we sought to tackle these issues by introducing two novel reconstruction methods: i) CaBBI (Calcium imaging Analysis using Biophysical models and Bayesian Inference), and ii) UFARSA (Ultra-fast Accurate Reconstruction of Spiking Activity). While CaBBI is a neuronal model-based method aiming at inferring the underlying biophysical quantities such as membrane potential (thus, spikes) and voltage-gated channels conductance, UFARSA is a heuristic method focusing on providing an ultra-fast, near-automatic reconstruction of spiking activities only. As the second goal, we sought to provide novel mechanistic insights into sparsification, whose underlying mechanisms are currently poorly understood. To this end, by integrating experimental findings, including only recently measured calcium imaging data covering the time course of sparsification, we present a dynamic systems modeling study of an in vivo developing neural network. With this approach, we address several aspects of sparsification such as i) the underlying mechanisms of cluster activity generation and its spatiotemporal characteristics, ii) the specific maturational processes mediating sparsification, and iii) the potential, developmental refinements in the networks information processing capabilities. Overall, this thesis not only provides novel tools of analyzing calcium imaging data in terms of the underlying biophysical quantities or ultra-fast near-automatic spiking activity reconstruction, but also presents how to analyze these data further with computational models in order to reveal the mechanisms of the sparsification, as a signature of cortex maturation.