Characterization of and compensation for the nonstationarity of spike shapes during physiological recordings

In a typical neurophysiological recording, the shape of the voltage trace due to an action potential depends, among other things, on the mutual geometry of the tissue and the electrode, and on the overall health of the preparation. These parameters change throughout the duration of the experiment, typically continuously and slowly, with occasional abrupt jumps. Here we propose a method to characterize such slow drifts and detect the abrupt changes by the estimation of a one-dimensional curved manifold, embedded in the largedimensional space of spike shapes, along which the data cluster. We model each event as the closest point on this manifold, which is the maximum-likelihood expectation of its spike shape, and a residual, which can be analyzed subsequently. We apply the method to both intraand extracellular recordings.