Kilosort: realtime spike-sorting for extracellular electrophysiology with hundreds of channels
read more
Citations
Single-trial neural dynamics are dominated by richly varied movements
Open Ephys: an open-source, plugin-based platform for multichannel electrophysiology.
Neuropixels 2.0: A miniaturized high-density probe for stable, long-term brain recordings
Genetic Dissection of Neural Circuits: A Decade of Progress.
A Fully Automated Approach to Spike Sorting
References
Matching pursuits with time-frequency dictionaries
Clustering by fast search and find of density peaks
An analysis of single-layer networks in unsupervised feature learning
Unsupervised spike detection and sorting with wavelets and superparamagnetic clustering
Large-scale recording of neuronal ensembles
Related Papers (5)
Spike sorting for large, dense electrode arrays
Fully integrated silicon probes for high-density recording of neural activity
Unsupervised spike detection and sorting with wavelets and superparamagnetic clustering
Frequently Asked Questions (11)
Q2. Why does Kilosort scale linearly with the number of recorded neurons?
The time taken to run Kilosort scales linearly with the number of recorded neurons, rather than the number of channels, due to the low-dimensional parametrization of template waveforms.
Q3. How many passes is the main loop alternating template matching and inference?
The main loop alternating template matching and inference is run until the cost function approaches convergence (typically less than six full passes through the data).
Q4. What is the oldest and reliable method for recording neural activity?
The oldest and most reliable method for recording neural activity involves lowering an electrode into the brain and recording the local electrical activity around the electrode tip.
Q5. How many spikes are allowed to be assigned to the same cluster?
The authors also anneal from small to large the ratio /λ, which controls the relative impact of the reconstruction term and amplitude bias term in equation 2; therefore, at the beginning of the optimization, spikes assigned to the same cluster are allowed to have more variable amplitudes.
Q6. How did the authors calculate the possible score after operator merges?
To estimate the best achievable score after operator merges, the authors took advantage ofthe ground truth data, and automatically merged together candidate clusters so as to greedily maximize their score.
Q7. What is the re-estimation of the running average?
Since firing rates vary over two orders of magnitude in typical recordings (from < 0.5 to 50 spikes/s), this adaptive running average procedure allows clusters with low firing rates to nonetheless average enough of their spikes to generate a smooth running-average template.
Q8. How many batches are used to obtain the templates?
After processing every hundred batches (or more, depending on their time length), the templates are obtained from the running average waveform
Q9. What is the generative model of the electrical recorded voltage?
To define a generative model of the electrical recorded voltage, the authors take advantage of the approximately linear summation of electrical potentials from different sources in the extracellular medium.
Q10. How did the authors avoid increasing the spike density at any location on the probe?
To avoid increasing the spike density at any location on the probe, the authors also subtracted off the denoised waveform from its original location.
Q11. What is the method for whitening the mean spike waveforms?
This method is based on the observation that any mean spike waveforms can be well explained by a singular value decomposition (SVD) decomposition of its spatiotemporal waveform, with as few as three components, but that the spatial and temporal components required can vary substantially between neurons (Fig. 2a).