Chronux: a platform for analyzing neural signals

Abstract

Chronux is an open-source software package developed for the analysis of neural data. The current version of Chronux includes software for signal processing of neural time-series data including several specialized mini-packages for spike-sorting, local regression, audio segmentation, and other data-analysis tasks typically encountered by a neuroscientist. Chronux is freely available along with user tutorials, sample data, and extensive documentation from http://chronux.org/.

Copyright 2010 Elsevier B.V. All rights reserved.

Figures

Figure 1. Typical workflow to analyze electrophysiological data

The data is filtered to obtain LFP and spiking activity, which are in turn pre-processed to remove artifacts and to exclude bad trials. The cleaned data-set is used for exploratory and confirmatory data analysis.

Figure 2. Spectrum computed using conventional methods and multi-taper methods

Spike-LFP coherence in area V4 in awake macaque during an attention guided saccade task. Shown in each graph are the coherences for two different conditions differing only in selective attention, along with jackknife estimates of the errors. The differences can only be seen in the Multitaper estimate on the left, smoothed with a 15 Hz bandwidth. The unsmoothed conventional estimate in the right panel, with a frequency resolution of 1 Hz, is not able to differentiate between the two conditions. Note also that the multi-taper method provides Jackknife based confidence intervals (figure courtesy Pascal Fries).

Figure 3. Spikesort Package

(a) Figure shows simulated raw voltage traces of spikes collected from a tetrode. Routines in the Spikesort mini-package allows users to sort such voltage traces to obtain spiking activity. (b) Noise on the electrode can jitter the exact time at which threshold crossing occurs and can be a significant source of variability for spikes from the same neuron. Routines in the Spikesort package allows users to remove such jitter and outliers for better clustering. (c) The algorithm implemented in Chronux deals with possibly non-Gaussian data (e.g., bursting neurons) by performing the sorting in two steps. The first step fits many local Gaussian spheres to the data to identify groups of spikes with similar shapes; which are combined into spike assignments in the second step.

Figure 4. LOCFIT package

(a) Regression fit using LOCFIT. A local regression fit, along with confidence intervals, where the independent variable was time and the dependent variable was simulated voltage from a cell. (b) Firing-rate estimate: LOCFIT can be used to estimate firing-rate given a sequence of spike times based on a local Poisson likelihood. Also shown are the LOCFIT computed 95% local confidence bands around the smoothed rate estimate. For comparison a histogram is also shown.

Figure 5. Spectrum

Comparison of a periodogram (black) and multitaper estimate (red or light) of a single trial local field potential measurement from macaque during a working memory task. This estimate used 9 tapers.

Figure 6. Spectrogram

Data from macaque monkey performing a working memory task: (a) Basline spectrogram of the LFP and (b) Spectrogram of the LFP during the memory period, which begins when the target is made visible at 2.5s, is plotted on a logarithmic scale. Sharp enhancement in high frequency power occurs during the working memory period.

Figure 7. Coherence

The spike-field coherence recorded from visual cortex of monkey, showing significant differences between the attended and unattended conditions. In addition to the coherence in the two conditions, we also show the 95% confidence bands computed using the Jackknife.