Spike sorting: new trends and challenges of the era of high-density probes

Recording from a large neuronal population of neurons is a crucial challenge to unravel how information is processed by the brain. In this review, we highlight the recent advances made in the field of “spike sorting”, which is arguably a very essential processing step to extract neuronal activity from extracellular recordings. We more specifically target the challenges faced by newly manufactured high-density multi-electrode array devices (HD-MEA), e.g. Neuropixels probes. Among them, we cover in depth the prominent problem of drifts (movements of the neurons with respect to the recording devices) and the current solutions to circumscribe it. In addition, we also review recent contributions making use of deep learning approaches for spike sorting, highlighting their advantages and disadvantages. Next, we highlight efforts and advances in unifying, validating, and benchmarking spike sorting tools. Finally, we discuss the spike sorting field in terms of its open and unsolved challenges, specifically regarding scalability and reproducibility. We conclude by providing our personal view on the future of spike sorting, calling for a community-based development and validation of spike sorting algorithms and fully automated, cloud-based spike sorting solutions for the neuroscience community.

How do spike collisions affect spike sorting performance?

Recently, a new generation of devices have been developed to record neural activity simultaneously from hundreds of electrodes with a very high spatial density, both for in vitro and in vivo applications. While these advances enable to record from many more cells, they also dramatically increase the amount overlapping "synchronous" spikes (colliding in space and/or in time), challenging the already complicated process of spike sorting (i.e. extracting isolated single-neuron activity from extracellular signals). In this work, we used synthetic ground-truth recordings to quantitatively benchmark the performance of state-of-the-art spike sorters focusing specifically on spike collisions. Our results show that while modern template-matching based algorithms are more accurate than density-based approaches, all methods, to some extent, failed to detect synchronous spike events of neurons with similar extracellular signals. Interestingly, the performance of the sorters is not largely affected by the the spiking activity in the recordings, with respect to average firing rates and spike-train correlation levels.

ProbeInterface: a unified framework for probe handling in extracellular electrophysiology

Recording neuronal activity with penetrating extracellular multi-channel electrode arrays, more commonly known as neural probes, is one of the most widespread approaches to probe neuronal activity. Despite a plethora of available extracellular probe designs, the time-consuming process of mapping of electrode channel order and relative geometries, as required by spike-sorting software is invariably left to the end-user. Consequently, this manual process is prone to mis-mapping mistakes, which in turn lead to undesirable spike-sorting errors and inefficiencies.Here we introduce ProbeInterface, an open-source project that aims to unify neural probe metadata descriptions by removing the manual step of probe mapping prior to spike-sorting for the analysis of extracellular neural recordings. ProbeInterface is first of all a Python API, which enables users to create and visualize probes and probe groups at any required complexity level. Second, ProbeInterface facilitates the generation of comprehensive wiring description ina reproducible fashion for any specific data-acquisition setup, which usually involves the use of a recording probe, a headstage, adapters, and an acquisition system. Third, we collaborate with probe manufacturers to compile an open library of available probes, which can be downloaded at run time using our Python API. Finally, with ProbeInterface we define a file format for probe handling which includes all necessary information for a FAIR probe description and is compatiblewith and complementary to other open standards in neuroscience.

Framework for automated sorting of neural spikes from Neuralynx-acquired tetrode recordings in freely-moving mice

Background Extracellular recording represents a crucial electrophysiological technique in neuroscience for studying the activity of single neurons and neuronal populations. The electrodes capture voltage traces that, with the help of analytical tools, reveal action potentials (‘spikes’) as well as local field potentials. The process of spike sorting is used for the extraction of action potentials generated by individual neurons. Until recently, spike sorting was performed with manual techniques, which are laborious and unreliable due to inherent operator bias. As neuroscientists add multiple electrodes to their probes, the high-density devices can record hundreds to thousands of neurons simultaneously, making the manual spike sorting process increasingly difficult. The advent of automated spike sorting software has offered a compelling solution to this issue and, in this study, we present a simple-to-execute framework for running an automated spike sorter. Methods Tetrode recordings of freely-moving mice are obtained from the CA1 region of the hippocampus as they navigate a linear track. Tetrode recordings are also acquired from the prelimbic cortex, a region of the medial prefrontal cortex, while the mice are tested in a T maze. All animals are implanted with custom-designed, 3D-printed microdrives that carry 16 electrodes, which are bundled in a 4-tetrode geometry. Results We provide an overview of a framework for analyzing single-unit data in which we have concatenated the acquisition system (Cheetah, Neuralynx) with analytical software (MATLAB) and an automated spike sorting pipeline (MountainSort). We give precise instructions on how to implement the different steps of the framework, as well as explanations of our design logic. We validate this framework by comparing manually-sorted spikes against automatically-sorted spikes, using neural recordings of the hippocampus and prelimbic cortex in freely-moving mice. Conclusions We have efficiently integrated the MountainSort spike sorter with Neuralynx-acquired neural recordings. Our framework is easy to implement and provides a high-throughput solution. We predict that within the broad field of bioelectronic medicine, those teams that incorporate high-density neural recording devices to their armamentarium might find our framework quite valuable as they expand their analytical footprint.

Evaluation and resolution of many challenges of neural spike-sorting: a new sorter

We evaluate existing spike sorters and present a new one that resolves many sorting challenges. The new sorter, called "full binary pursuit" or FBP, comprises multiple steps. First, it thresholds and clusters to identify the waveforms of all unique neurons in the recording. Second, it uses greedy binary pursuit to optimally assign all the spike events in the original voltages to separable neurons. Third, it resolves spike events that are described more accurately as the superposition of spikes from two other neurons. Fourth, it resolves situations where the recorded neurons drift in amplitude or across electrode contacts during a long recording session. Comparison with other sorters on ground-truth datasets reveals many of the failure modes of spike sorting. We examine overall spike sorter performance in ground-truth datasets and suggest post-sorting analyses that can improve the veracity of neural analyses by minimizing the intrusion of failure modes into analysis and interpretation of neural data. Our analysis reveals the tradeoff between the number of channels a sorter can process, speed of sorting, and some of the failure modes of spike sorting. FBP works best on data from 32 channels or fewer. It trades speed and number of channels for avoidance of specific failure modes that would be challenges for some use cases. We conclude that all spike sorting algorithms studied have advantages and shortcomings, and the appropriate use of a spike sorter requires a detailed assessment of the data being sorted and the experimental goals for analyses.

Characteristic of Neural Signal Feature for Spike Sorting and Detection

Neural spike plays an important role in understanding brain activities, and in neural spike sorting, the features of signal are of great importance. This paper aims to have a review on features used to discriminate different originated spikes. The features are divided into three categories: features in the time domain, features in the transformation domain, and features of dimensional reduction. For each kind of feature, the basic principle, advantages, and disadvantages are described and discussed. Results showed that features in the time domain are suitable for on-chip or real-time spike sorting, while features in the transformation domain can be used in offline spike sorting aiming at high performance. For features of dimensional reduction, it makes a large number of features available in spike sorting. In conclusion, researchers need to determine features by balancing the minimization of calculation complexity and maximizing sorting performance according to different occasions and demands. Expectations are also made for future directions of spike feature studies. The article may guide both the physiologists who want to determine features in neural spike sorting and researchers who want to work on feature extracting algorithms further to achieve better performance in experimental challenges.