Multi-neuronal activity and functional connectivity in cell assemblies Yasser Roudi 1,2 , Benjamin Dunn 1 and John Hertz 3,2 Our ability to collect large amounts of data from many cells has been paralleled by the development of powerful statistical models for extracting information from this data. Here we discuss how the activity of cell assemblies can be analyzed using these models, focusing on the generalized linear models and the maximum entropy models and describing a number of recent studies that employ these tools for analyzing multi- neuronal activity. We show results from simulations comparing inferred functional connectivity, pairwise correlations and the real synaptic connections in simulated networks demonstrating the power of statistical models in inferring functional connectivity. Further development of network reconstruction techniques based on statistical models should lead to more powerful methods of understanding functional anatomy of cell assemblies. Addresses 1 Kavli Institute & Centre for Neural Computation, NTNU, Trondheim, Norway 2 Nordita, KTH Royal Institute of Technology and Stockholm University, Stockholm, Sweden 3 Niels Bohr Institute, Copenhagen, Denmark Corresponding author: Roudi, Yasser (yasser.roudi@ntnu.no) Current Opinion in Neurobiology 2015, 32:38–44 This review comes from a themed issue on Large-scale recording technology Edited by Francesco Battaglia and Mark J Schnitzer http://dx.doi.org/10.1016/j.conb.2014.10.011 0959-4388/# 2014 Elsevier Ltd. All right reserved. Introduction In lower species, single neurons or small circuits with stereotyped connectivity patterns are studied as compu- tational building blocks of the nervous system. In higher species, such as mammals, on the other hand, populations of neurons, or cell assemblies, are probably the closest thing to a computational unit [1]. A familiar example is the Hebbian cell assembly [2]: a group of neurons with stronger connections between the cells within the group than with other cells. The stronger connections between the neurons in the Hebbian assembly leads to the attractor dynamics that is believed to underlie a variety of neuronal compu- tations [3]. Other examples of cell assemblies include groups of neurons that share functional similarities, such as color, form and motion selective cells in primary visual areas [4], or the barrels in the rat barrel cortex [5]. Func- tional cells assemblies also exist in higher cortical areas. An example is that of grid cells in the medial entorhinal cortex [6]. As an animal runs in a two-dimensional environment, each grid cell fires maximally at locations that form a hexagonal pattern. Grid cells form a functional cell assem- bly and are coupled to cells in the same anatomical location that belong to other assemblies, for example, border cells [7] and head directional cells [8]. To understand compu- tation in the mammalian nervous system, one has to characterize these assemblies and their relationship to each other and to identify the anatomical and molecular features associated with specific assemblies. Although the theoretical concept of cell assemblies is not new, tools for analyzing them have only recently emerged in systems neuroscience. Experimentalists can now record the activity of many cells at the same time, and the spatial and temporal resolution with which these recordings can be done is increasing rapidly [9,10]. With new recording technology, even areas previously inac- cessible to simultaneous multi-cell recording are becom- ing available. In addition, optogenetic [11] and other molecular and genetic techniques [12,13] now allow experimentalists to stimulate specific kinds of cells during their recordings. All these advances have shifted the focus of efforts to understand neural computation from single- neuron recordings to simultaneous recordings of many neurons. In parallel, there has been significant progress on theoretical and computational tools for analyzing such recordings. Although, so far, these methods have been applied almost exclusively to data from sensory or motor areas, we can anticipate their exploitation in higher cor- tical areas, leading to new ways of thinking about infor- mation processing at the population level. In this paper, we review the main modern approaches for modeling multi-unit recordings and discuss future avenues that can be explored using these methods. Statistical modeling Understanding a complex system is achieved through models, and the high variability of neuronal data requires that these models be statistical ones. Here, we describe how to build statistical models of multi-neuronal activity and show, using several examples, how they can help us understand the computational and physiological proper- ties of cells assemblies, as well as the relationship be- tween them. Available online at www.sciencedirect.com ScienceDirect Current Opinion in Neurobiology 2015, 32:38–44 www.sciencedirect.com