Finding the right measure: New study allows a more reliable measurement of neuronal interactions
Click to enlarge. See image caption below.
To answer this question, the crucial first step is to collect experimental data and to measure the degree of correlation with high precision. Such data sets are very large and complex, and detecting correlations is beyond the abilities of the human eye inspecting these data. Therefore, powerful statistical procedures are used as analytical tools. However, such methods often assume that neural activity can be well described by a Poisson process, meaning that the occurrence of a spike is equally likely for all instants of time. This view, however, is rather far away from the biological reality of a nerve cell. For instance, a neuron needs to recover after it has generated a spike, and this imposes a certain dead time where no other spike can be generated.
Does such innocent-looking deviation from the ideal description of neural activity as a Poisson process impair the measurement of spike correlations? Imke Reimer and her colleagues from the Bernstein Center Freiburg and the Institute for Frontier Areas of Psychology and Mental Health in Freiburg addressed this question in an article just published in the Journal of Neuroscience Methods. In their study, the scientists focused on spike correlations among three or more neurons, dubbed “higher-order” correlations in the jargon used by neuro-statisticians.
In their study, the researchers proposed a new method to mimic populations of spiking neurons with higher-order correlations, equipped with properties much closer to biology than the commonly used Poisson process. Using this method in computer simulations, they were able to create artificial data with known properties, which allowed them to test and calibrate their methods to uncover higher-order correlations. For this detection, they relied on a procedure called “empirical de-Poissonization” (further information on this method can be found in Ehm et al., Electronic Journal of Statistics, 2007).
The results obtained by this procedure led the scientists from Freiburg to propose an elaborate strategy to assess the reliability of an analysis of real data. Doing so is of outstanding importance, because it helps to avoid the misinterpretation of empirical data recorded in electrophysiological experiments, which could otherwise lead to false conclusions about the role of correlated activity in the brain.
Image caption:
The paper proposes a new method to generate artificial neuronal signals that match the statistical properties of biological neurons (blue) and, at the same time, imitate the activity of hypothetical groups of correlated neurons (coloured spikes). Surrogate data are instrumental for the verification of statistical methods to uncover such higher-order interactions in real spike trains.
Imke C.G. Reimer, Benjamin Staude, Werner Ehm, and Stefan Rotter (2012) Modeling and analyzing higher-order correlations in non-Poissonian spike trains. Journal of Neuroscience Methods 208 (1), 18–33.