Making neurophysiological data analysis reproducible: why and how?

Reproducible data analysis is an approach aiming at complementing classical printed scientific articles with everything required to independently reproduce the results they present. "Everything" covers here: the data, the computer codes and a precise description of how the code was applied to the data. A brief history of this approach is presented first, starting with what economists have been calling replication since the early eighties to end with what is now called reproducible research in computational data analysis oriented fields like statistics and signal processing. Since efficient tools are instrumental for a routine implementation of these approaches, a description of some of the available ones is presented next. A toy example demonstrates then the use of two open source software programs for reproducible data analysis: the "Sweave family" and the org-mode of emacs. The former is bound to R while the latter can be used with R, Matlab, Python and many more "generalist" data processing software. Both solutions can be used with Unix-like, Windows and Mac families of operating systems. It is argued that neuroscientists could communicate much more efficiently their results by adopting the reproducible research paradigm from their lab books all the way to their articles, thesis and books.

Efficient spike-sorting of multi-state neurons using inter-spike intervals information.

We demonstrate the efficacy of a new spike-sorting method based on a Markov chain Monte Carlo (MCMC) algorithm by applying it to real data recorded from Purkinje cells (PCs) in young rat cerebellar slices. This algorithm is unique in its capability to estimate and make use of the firing statistics as well as the spike amplitude dynamics of the recorded neurons. PCs exhibit multiple discharge states, giving rise to multi-modal inter-spike interval (ISI) histograms and to correlations between successive ISIs. The amplitude of the spikes generated by a PC in an "active" state decreases, a feature typical of many neurons from both vertebrates and invertebrates. These two features constitute a major and recurrent problem for all the presently available spike-sorting methods. We first show that a hidden Markov model with three log-normal states provides a flexible and satisfying description of the complex firing of single PCs. We then incorporate this model into our previous MCMC based spike-sorting algorithm [Pouzat C, Delescluse M, Viot P, Diebolt J. Improved spike-sorting by modeling firing statistics and burst-dependent spike amplitude attenuation: a Markov chain Monte Carlo approach. J Neurophysiol 2004;91:2910-28] and test this new algorithm on multi-unit recordings of bursting PCs. We show that our method successfully classifies the bursty spike trains fired by PCs by using an independent single unit recording from a patch-clamp pipette.

Improved spike-sorting by modeling firing statistics and burst-dependent spike amplitude attenuation: a Markov chain Monte Carlo approach.

Spike-sorting techniques attempt to classify a series of noisy electrical waveforms according to the identity of the neurons that generated them. Existing techniques perform this classification ignoring several properties of actual neurons that can ultimately improve classification performance. In this study, we propose a more realistic spike train generation model. It incorporates both a description of "nontrivial" (i.e., non-Poisson) neuronal discharge statistics and a description of spike waveform dynamics (e.g., the events amplitude decays for short interspike intervals). We show that this spike train generation model is analogous to a one-dimensional Potts spin-glass model. We can therefore tailor to our particular case the computational methods that have been developed in fields where Potts models are extensively used, including statistical physics and image restoration. These methods are based on the construction of a Markov chain in the space of model parameters and spike train configurations, where a configuration is defined by specifying a neuron of origin for each spike. This Markov chain is built such that its unique stationary density is the posterior density of model parameters and configurations given the observed data. A Monte Carlo simulation of the Markov chain is then used to estimate the posterior density. We illustrate the way to build the transition matrix of the Markov chain with a simple, but realistic, model for data generation. We use simulated data to illustrate the performance of the method and to show that this approach can easily cope with neurons firing doublets of spikes and/or generating spikes with highly dynamic waveforms. The method cannot automatically find the "correct" number of neurons in the data. User input is required for this important problem and we illustrate how this can be done. We finally discuss further developments of the method.