Neural activity classification with machine learning models trained on interspike interval series data
share
The flow of information through the brain is reflected by the activity patterns of neural cells. Indeed, these firing patterns are widely used as input data to predictive models that relate stimuli and animal behavior to the activity of a population of neurons. However, relatively little attention was paid to single neuron spike trains as predictors of cell or network properties in the brain. In this work, we introduce an approach to neuronal spike train data mining which enables effective classification and clustering of neuron types and network activity states based on single-cell spiking patterns. This approach is centered around applying state-of-the-art time series classification/clustering methods to sequences of interspike intervals recorded from single neurons. We demonstrate good performance of these methods in tasks involving classification of neuron type (e.g. excitatory vs. inhibitory cells) and/or neural circuit activity state (e.g. awake vs. REM sleep vs. nonREM sleep states) on an open-access cortical spiking activity dataset.
READ FULL TEXT