Validating silicon polytrodes with paired juxtacellular recordings: method and dataset

Abstract

Recording in vivo from the same neuron with two different methods is difficult. It requires blindly moving each probe to within ∼100 μm of one another and for this reason such “dual-recordings” are rare. However, comparing the signals measured by different techniques is necessary to understand what they measure. We developed a method to precisely align the axes of two manipulators and used it to gather a “ground truth” dataset for dense extracellular polytrodes.

A Fully Automated Approach to Spike Sorting

“…Setting aside the issue of human intervention, we compared MountainSort (MS) with two other spike sorting packages: KiloSort (KS) (Pachitariu et al, 2016) and Spyking Circus (SC) (Yger et al, 2016). The three algorithms were applied to (a) real data from our laboratory (the tetrode dataset described above), (b) a publicly available extracellular dataset with known ground truth (128-channel silicon polytrode together with a juxtacellular ground-truth measurement) (Neto et al, 2016), and (c) simulated data obtained from superimposing synthetic waveforms on background signal taken from a real dataset. Each of the three software packages has parameters that can be modified to optimize performance for different applications.…”

“…We then applied MS, KS, and SC to a publicly available 128-channel dataset with independent juxtacellular firing information for one of the cells (Neto et al, 2016). This dataset is one of ten datasets in the repository exhibiting varying levels of sorting difficulty.…”

“…This is 30 times real time. The 128-channel publicly available dataset with juxtacellular ground truth (Neto et al, 2016) was sorted in 249 seconds (including 128 seconds preprocessing) using 40 threads. This represents 2.5 times real time.…”

Improving data quality in neuronal population recordings

“…While invertebrate and in vitro preparations have provided some data on which to test algorithms 40–42 , the noise conditions and nonstationarity found in mammalian systems in vivo are substantially more challenging. The difficulty of obtaining ground truth data in vivo has made such data very rare 8, 36, 43 (https://crcns.org/data-sets/hc/hc-1). The data that is available, however, suggests that error rates for semi-automatic clustering with tetrodes can be of the order 5-10%, but the error rates of purely manual cluster cutting may be substantially higher 36 .…”

Recent progress in multi-electrode spike sorting methods

“…recordings where at least one cell is recorded with another technique, so that we know when the spikes occur. These data are essential to test spike sorting algorithms (Neto et al (2016)). …”