Neural data science: accelerating the experiment-analysis-theory cycle in large-scale neuroscience

Abstract

Modern large-scale multineuronal recording methodologies, including multielectrode arrays, calcium imaging, and optogenetic techniques, produce single-neuron resolution data of a magnitude and precision that were the realm of science fiction twenty years ago. The major bottlenecks in systems and circuit neuroscience no longer lie in simply collecting data from large neural populations, but also in understanding this data: developing novel scientific questions, with corresponding analysis techniques and experimental designs to fully harness these new capabilities and meaningfully interrogate these questions. Advances in methods for signal processing, network analysis, dimensionality reduction, and optimal control -developed in lockstep with advances in experimental neurotechnology --promise major breakthroughs in multiple fundamental neuroscience problems. These trends are clear in a broad array of subfields of modern neuroscience; this review focuses on recent advances in methods for analyzing neural time-series data with single-neuronal precision. Topics reviewed herein are indicated in black.. CC-BY-NC-ND 4.0 International license peer-reviewed) is the author/funder. It is made available under aThe copyright holder for this preprint (which was not . http://dx.doi.org/10.1101/196949 doi: bioRxiv preprint first posted online Oct. 2, 2017; I. High-throughput neural signal processing methodsNeuroscientists have long dreamed of recording from many thousands of neurons simultaneously. This goal is the major motivation of the BRAIN initiative and related efforts, and with new calcium imaging methods and large-scale multielectrode array (MEA) devices, this dream is quickly becoming a reality. But now a major bottleneck exists. Cutting-edge calcium imaging methods and MEAs output data at rates on the order of terabytes/hour , and data rates continue to increase. At these huge rates processing and even storing the data is challenging (Freeman et al 2014 Nature Methods), let alone optimally extracting all the useful information in these data streams; without the right analytical technology, we will never unlock the true potential of these experimental advances. Calcium imagingCalcium imaging has become the dominant method for recording from large populations of neurons, due to several well-known advantages: calcium imaging offers cell-type specificity and can be coupled easily with a variety of genetic tools; imaging approaches can be less invasive and damaging to brain tissue than inserting an MEA; calcium imaging has proven scalability to record simultaneously from O(10 4 ) neurons in vivo (over an order of magnitude larger than achieved by an MEA to date); and finally, imaging approaches enable significantly greater experimental design flexibility than MEAs in terms of which subsets of neurons in the imaging volume are interrogated at which times, and how many pixels are assigned to each neuron (we expand on the importance of this point below). At the same time, calcium imaging suffers from some clear disadvan...

A roadmap to integrate astrocytes into Systems Neuroscience

“…Alternatively, shot‐gun statistics unravels network connectivity information from recording at only 10% of the neurons at a given time, thus simplifying the experimental load of large‐scale recordings (Soudry et al, ). Data‐sharing and collaborative solutions have been proposed as well to manage the surge of data (Paninski & Cunningham, ).…”

“…For encoding, generalized linear models (GLMs), a generalization of multiple linear regression, regress neuronal activity against behavioral variables to determine the set of variables that explain more neuronal activity (Aljadeff et al, ; Nogueira et al, ). Decoding techniques, typically linear classifiers (Arandia‐Romero, Nogueira, Mochol, & Moreno‐Bote, ; Quian Quiroga & Panzeri, ), as well as more recent artificial neural networks (ANNs; Paninski & Cunningham, ) are used to predict, trial‐by‐trial, values of behavioral variables from neuronal activity, either using single neuronal activity or the individual activity of large neuronal populations recorded from multielectrode‐arrays or Ca 2+ imaging. These methods are supervised machine learning tools because both behavioral and neuronal variables are preselected and labeled.…”

“…Thus, dimensionality reduction of single‐astrocytes may help to reveal and select dimensions, that is, the minimum number of ROIs (e.g., 5–10 from up to 200 original ones), in which fluctuations are more pronounced and meaningful, thus paving the way for population analyses, which will require the simplification of Ca 2+ signals per astrocyte with the minimal loss of relevant information. Linear methods for dimensionality reduction that can be used in astrocytes include simple principal component analysis (PCA), the prime linear method (Cunningham & Yu, ), as well as factor analysis, as used with neuronal Ca 2+ (Paninski & Cunningham, ).…”

“…Nonlinear methods such as ANNs are increasingly being used to replace stages in signal processing and analysis in neuronal populations, as well as a method for dimensionality reduction (Paninski & Cunningham, ). Thus, ANNs could a priori uncover latent variables that best account for Ca 2+ data from astrocyte mini‐circuits, and are nonlinearly related.…”

“…Thus, ANNs could a priori uncover latent variables that best account for Ca 2+ data from astrocyte mini‐circuits, and are nonlinearly related. Current ANNs appear well‐suited to extract latent variables from Ca 2+ imaging of large populations of neurons (Paninski & Cunningham, ), and their application to multidimensional astrocytic Ca 2+ data should be explored. Conversely, ANNs can be also used as generative models, that is, models that infer classes of inputs from a low number of latent variables (Dosovitskiy, Springenberg, & Brox, ).…”