spike-sorting

Code to perform spike sorting on a multi-electrode array using one of three methods:

• Single-electrode sorting via clustering: Clusters snippets of data with high activity to produce spike and waveform estimates for a single electrode.

• Optimization-based single-electrode sorting: Estimates the spike locations and the corresponding waveforms for a single electrode by optimizing the fit to the data.

• Optimization-based multi-electrode sorting: Estimates the spike locations for each cell over the whole array, as well as the corresponding waveforms, by optimizing the fit to the data.

1. Single-electrode sorting via clustering

Locates snippets of data with high activity, upsamples them and clusters them.

Snippet extraction

The data is segmented into snippets, which are then upsampled and aligned.

segment_data.py

Segments data by thresholding the energy over a windows of a certain length. The threshold is set so that a certain fraction of the data is selected.

• segment_data(data, wl, max_fraction, min_fraction): A point i is selected if the energy within a window of length wl centered at i is above a certain threshold. The threshold is set so that the fraction of data that is selected is between min_fraction and max_fraction

locate_snippets.py

Locates snippets by finding local extrema in the data

• locate_snippets(data, ind_ini, ind_end, wl): Locates points that have the largest magnitude within within a window of length wl. The search is restricted to the segments indexed by ind_ini and ind_end.

upsample_snippets.py

Upsamples snippets and centers them.

• upsample_snippets(data, snippet_indices, wl, uf): Snippets located at snippet_indices are separated into two types: those that first have a maximum and then a minimum (corresponding to soma-dendritic spikes) and vice versa (axonal spikes). The snippets are upsampled by a factor uf and then aligned so that their minimum coincides.

K-means clustering

cluster_snippets.py:

• cluster_snippets(snippets, locations, k, data, uf, wl, p=0, average=True, save_file="")

Clusters snippets produced by upsample_snippets using k-means, optionally after projecting onto the first p principal components.

Additional functions:

Data preprocessing

preprocess_data.py

Applies a high-pass filter and normalizes the amplitude of the electrode data.

• filter_data(data, params): Applies a high-pass Butterworth filter.

• preprocess_data(data, params=None): Calls filter_data and then normalizes the data.

Evaluation

2. Multi-electrode sorting

Detect pattern of multi-electrode patterns, use non-negative matrix factorization