Test of spike-sorting algorithms on the basis of simulated network data

doi:10.1016/S0925-2312(02)00432-0

Neurocomputing

Volumes 44–46, June 2002, Pages 1119-1126

https://doi.org/10.1016/S0925-2312(02)00432-0 Get rights and content

Abstract

Results of spike-sorting algorithms are usually compared with recorded signals which themselves underly interpretations, distortions and errors. Our approach is to provide and compare physiological extracellular potential data by a realistic cortical network simulation. For this purpose, we utilize the neural simulator GENESIS and simulate a region of rat hippocampus containing 90 cells. We are able to “record” simulated extracellular potentials from “virtual electrodes” and produce test data closely resembling multisite neuronal recordings. Our realisitic, artificial data are complex and almost natural in appearance; however, current spike detection schemes appear unable to reliably detect all spikes produced.

Introduction

Progress in microtechnology development by the European VSAMUEL consortium [5] following the track started earlier at the University of Michigan [2], recently achieved multisite recording probes with 32 electrode sites on multiple purpose silicon probes. This opens the door to acquire neuronal signals which may contain spike trains from hundreds of cells [16]. Obviously, the amount of data acquired in a single experiment requires some type of automation to assign spikes to individual cells. This spike sorting is currently done by several more or less standard methods. However, a rigid assessment of their quality is needed.

Until now, results of spike-sorting algorithms have been compared with recorded signals which themselves underly interpretations, distortions, and errors. An alternative is, to run tests on artificial data generated by adding spike snippets to noisy signals [3]. This method has the advantage of control over data, but is paid for with unphysiological data. Our approach is to mimic physiological extracellular potential data in a biologically realistic network simulation. For this purpose, we utilize the freely available neural simulator GENESIS [1] to simulate a small network of 90 cortical cells and use their output for assessment within two automated spike detection algorithms [10], [17].

Section snippets

Methods

A region of cortex is represented by 72 CA3 pyramidal cells (PYR), randomly spaced at 35 up to $45 μm$ in each direction, and 18 interspersed inhibitory interneurons. The interneurons are divided into nine feedforward and nine feedback interneurons. The only difference between the two inhibitory cell types is their pattern of connectivity. Neurons of one group are 75– $85 μm$ in x-direction and 155– $165 μm$ in y-direction apart from each other, thus alternating in y-direction. Additionally, there is a

Results and discussions

Whereas our small cortical model awaits its physiological validation by real brain recordings, the simulated extracellular, multiunit signals resemble closely experimental multisite recordings taken with silicon probes by [16]. Specifically, a bell-shaped distribution in potential amplitude along a linear site array (Fig. 3) can be found corresponding to real recordings as well. The middle electrode (i.e. GENESIS efield-object No. 4 of each array, see Fig. 2) is located close to simulated

Acknowledgements

This work was supported in part by the EU Grant IST-1999-10079.

References (17)

G. Buzsaki et al.
Multisite recording of brain field potentials and unit-activity in freely moving rats
J. Neurosci. Meth.
(1989)
B.H. McCormick et al.
Geometric modeling of local cortical networks
Neurocomputing
(2000)
B. Weber et al.
Handling large files of multisite microelectrode recordings for the European VSAMUEL consortium
Neurocomputing
(2001)
J.M. Bower et al.
The Book of GENESIS
(1998)
R. Chandra et al.
Detection, classification, and superposition resolution of action potentials in multiunit single-channel recordings by an on-line real-time neural network
IEEE Trans. Biomed. Eng.
(1997)
F. Gabbiani et al.
Synaptic integration in a model of cerebellar granule cells
J. Neurophysiol.
(1994)
U.G. Hofmann, E. De Schutter, et al., On the Design of Multi-site Microelectrodes for Neuronal Recordings. in MICRO.tec...
E.R. Kandel et al.
Principles of Neural Science
(1991)

There are more references available in the full text version of this article.

Cited by (16)

Neural networks: An overview of early research, current frameworks and new challenges
2016, Neurocomputing
Citation Excerpt :
Thus, it is preferable to include the website of each simulator, where it is possible to find the specific details and always in an updated form. There are many other programs and frameworks, either of general purpose or that simulate functions or neural structures described in the literature, for instance: IQR [180], NeuroSpaces [181], NNET [182], NeuralSyns [183], NEUVISION [184], NeuroWeb [185], RSNNS [186], and See [187], or of particular models or levels as referenced in [188–192]. There is no specific simulator that is currently being used by the whole community (since some different approaches are more suitable than others, depending on the research task being addressed).
This paper presents a comprehensive overview of modelling, simulation and implementation of neural networks, taking into account that two aims have emerged in this area: the improvement of our understanding of the behaviour of the nervous system and the need to find inspiration from it to build systems with the advantages provided by nature to perform certain relevant tasks. The development and evolution of different topics related to neural networks is described (simulators, implementations, and real-world applications) showing that the field has acquired maturity and consolidation, proven by its competitiveness in solving real-world problems. The paper also shows how, over time, artificial neural networks have contributed to fundamental concepts at the birth and development of other disciplines such as Computational Neuroscience, Neuro-engineering, Computational Intelligence and Machine Learning. A better understanding of the human brain is considered one of the challenges of this century, and to achieve it, as this paper goes on to describe, several important national and multinational projects and initiatives are marking the way to follow in neural-network research.
ViSAPy: A Python tool for biophysics-based generation of virtual spiking activity for evaluation of spike-sorting algorithms
2015, Journal of Neuroscience Methods
Citation Excerpt :
Such ground-truth data sets, analogous to well-established benchmarking data sets in computer science (Hockney, 1996), can in principle be obtained experimentally by means of joint intracellular and extracellular recordings of action potentials (Henze et al., 2000; Harris et al., 2000), but such double recordings are difficult to do, and in practice limited to a modest number of neurons. A natural alternative is to use synthetic benchmark data, either constructed from experimentally measured spike waveforms (Wood et al., 2004; Thorbergsson et al., 2010), biophysical forward modeling (Menne et al., 2002; Eaton and Henriquez, 2005; Smith and Mtetwa, 2007; Franke et al., 2010; Thorbergsson et al., 2002), or a combination of both (Martinez et al., 2009; Camuñas Mesa and Quiroga, 2013). A limitation with use of experimentally recorded waveform templates is that not only the spike amplitude, but also the spike shape, depend on electrode position.
New, silicon-based multielectrodes comprising hundreds or more electrode contacts offer the possibility to record spike trains from thousands of neurons simultaneously. This potential cannot be realized unless accurate, reliable automated methods for spike sorting are developed, in turn requiring benchmarking data sets with known ground-truth spike times.
We here present a general simulation tool for computing benchmarking data for evaluation of spike-sorting algorithms entitled ViSAPy (Virtual Spiking Activity in Python). The tool is based on a well-established biophysical forward-modeling scheme and is implemented as a Python package built on top of the neuronal simulator NEURON and the Python tool LFPy.
ViSAPy allows for arbitrary combinations of multicompartmental neuron models and geometries of recording multielectrodes. Three example benchmarking data sets are generated, i.e., tetrode and polytrode data mimicking in vivo cortical recordings and microelectrode array (MEA) recordings of in vitro activity in salamander retinas. The synthesized example benchmarking data mimics salient features of typical experimental recordings, for example, spike waveforms depending on interspike interval.
ViSAPy goes beyond existing methods as it includes biologically realistic model noise, synaptic activation by recurrent spiking networks, finite-sized electrode contacts, and allows for inhomogeneous electrical conductivities. ViSAPy is optimized to allow for generation of long time series of benchmarking data, spanning minutes of biological time, by parallel execution on multi-core computers.
ViSAPy is an open-ended tool as it can be generalized to produce benchmarking data or arbitrary recording-electrode geometries and with various levels of complexity.
Towards reliable spike-train recordings from thousands of neurons with multielectrodes
2012, Current Opinion in Neurobiology
Citation Excerpt :
Such dual recordings are difficult to do, however, and the yield in terms of the number of ground-truth spike trains will by nature be limited. An attractive alternative has thus been to use synthetic test data obtained from modeling [32,34–38]. In one approach synthetic data has been constructed based on extracellular spike shapes extracted from real data, superposed in a stochastic manner to incorporate a plausible underlying spike-train statistics [10,32,34,36].
The new generation of silicon-based multielectrodes comprising hundreds or more electrode contacts offers unprecedented possibilities for simultaneous recordings of spike trains from thousands of neurons. Such data will not only be invaluable for finding out how neural networks in the brain work, but will likely be important also for neural prosthesis applications. This opportunity can only be realized if efficient, accurate and validated methods for automatic spike sorting are provided. In this review we describe some of the challenges that must be met to achieve this goal, and in particular argue for the critical need of realistic model data to be used as ground truth in the validation of spike-sorting algorithms.
New approaches to eliminating common-noise artifacts in recordings from intracortical microelectrode arrays: Inter-electrode correlation and virtual referencing
2009, Journal of Neuroscience Methods
Intracortical microelectrode arrays record multi-unit extracellular activity for neurophysiology studies and for brain–machine interface applications. The common first step is neural spike-detection; a process complicated by common-noise signals from motion artifacts, electromyographic activity, and electric field pickup, especially in awake/behaving subjects. Often common-noise spikes are very similar to neural spikes in their magnitude, spectral, and temporal features. Provided sufficient spacing exists between electrodes of the array, a local neural spike is rarely recorded on multiple electrodes simultaneously. This is not true for distant common-noise sources. Two new techniques compatible with standard spike-detection schemes are introduced and evaluated. The first method, virtual referencing (VR), takes the average recording from all functional electrodes in the array (represents the signal from a virtual-electrode at the array's center) and subtracts it from the test electrode signal. The second method, inter-electrode correlation (IEC), computes a correlation coefficient between threshold exceeding candidate spike segments on the test electrode and concurrent segments from remaining electrodes. When sufficient correlation is detected, the candidate spike is rejected as originating from a distant common-noise source. The performance of these algorithms was compared with standard thresholding and differential referencing approaches using neural recordings from un-anaesthetized rats. By evaluating characteristics of mean-spike waveforms generated by each method under different levels of common-noise, it was found that IEC consistently offered the most robust means of neural spike-detection. Furthermore, IEC's rejection of supra-threshold events not likely originating from local neurons significantly reduces data handling for downstream spike sorting and processing operations.
Depolarizing, GABA-mediated synaptic responses and their possible role in epileptiform events; Simulation studies
2007, Neurocomputing
Synaptic transmission controlled by the release of the inhibitory neurotransmitter GABA is generally considered to be the mechanism by which glutamate-mediated excitation is kept under control in the human hippocampus and neocortex. Until recently, epilepsy was thought of as a simple imbalance of neuronal excitation and inhibition. This theory of epileptogenesis has been challenged by the finding that GABA does not always inhibit neuronal activity. Although in adult brain GABA usually induces hyperpolarization of cell membranes, in juvenile brain GABA is depolarizing, bringing the neuronal membrane closer to firing threshold, often enabling action potentials to be triggered. Somewhat surprisingly, GABA-mediated synaptic responses in adult brain tissue can sometimes be excitatory, too.
Cohen et al. [I. Cohen, V. Navarro, S. Clemenceau, M. Baulac, R. Miles, On the origin of interictal activity in human temporal lobe epilepsy in vitro, Science 298 (2002) 1418–1421] discovered a subpopulation of excitatory projection cells that exhibited depolarizing GABA responses in slices from epileptic adult subiculum. While most subicular pyramidal cells displayed hyperpolarized behavior in response to GABA, some actually fired bursts of action potentials, in synchrony with the interneurons; the GABA released by the interneurons was only depolarizing for this subset of excitatory cells.
These results suggest that it is the interaction between excitatory projection cells depolarized by GABA and interneurons that initiates epileptiform events, at least in subiculum. Interneurons usually seem to provide an inhibitory shield around excitatory neurons. However, when connected to projection cells responding ‘abnormally’ to GABA with depolarization, they may promote paroxysmal synchronizations.
A summary of some recent theories and results is presented on possible causes and effects of depolarizing, GABA-mediated synaptic responses in the cerebrum. Special attention is focused on the hippocampal/parahippocampal formation, especially the subicular complex. The subiculum seems to be of particular interest because of its strategic output location to neocortex, and because of the spontaneous, interictal-like activity observed almost exclusively there in slices from patients suffering mesial temporal lobe epilepsy.
Simulation results follow, starting from a commonly known GENESIS computer model of a small part of CA3. This ‘CA3’ was re-modeled to more closely resemble a small, ‘subiculum-like’ structure, with a complement of fast-spiking interneurons and three types of projection cells: ‘strong-bursting’, ‘weak-bursting’, and ‘regular-spiking’. Parametric studies of the effects of increasing $E_{{GABA}_{A}},$ the GABA reversal potential of certain GABA_A receptors, simulate the sometimes excitatory impact of GABAergic signaling. The effects of GABA_B receptor impairment in this setting are also briefly considered. Results presented here reinforce experimental evidence that the subiculum has “the right stuff” to play a significant role in epileptiform events.
A tool for synthesizing spike trains with realistic interference
2007, Journal of Neuroscience Methods
Spike detection and spike sorting techniques are often difficult to assess because of the lack of ground truth data (i.e., spike timings for each neuron). This is particularly important for in vitro recordings where the signal to noise ratio is poor (as is the case for multi-electrode arrays at the bottom of a cell culture dish). We present an analysis of the transmission of intracellular signals from neurons to an extracellular electrode, and a set of MATLAB functions based on this analysis. These produce realistic signals from neighboring neurons as well as interference from more distant neurons, and Gaussian noise. They thus generate realistic but controllable synthetic signals (for which the ground truth is known) for assessing spike detection and spike sorting techniques. They can also be used to generate realistic (non-Gaussian) background noise. We use signals generated in this way to compare two automated spike-sorting techniques. The software is available freely on the web.

View all citing articles on Scopus

View full text

Test of spike-sorting algorithms on the basis of simulated network data

Abstract

Introduction

Section snippets

Methods

Results and discussions

Acknowledgements

J. Neurosci. Meth.

Neurocomputing

Neurocomputing

The Book of GENESIS

Detection, classification, and superposition resolution of action potentials in multiunit single-channel recordings by an on-line real-time neural network

IEEE Trans. Biomed. Eng.

Synaptic integration in a model of cerebellar granule cells

J. Neurophysiol.

Principles of Neural Science