The Oxford Mouse Polysomnography Benchmark Data Set

This archive contains electrophysiological recordings from freely behaving mice, with each 4-second epoch having been annotated with the corresponding vigilance state by multiple sleep experts. The recordings are either 12 or 24 hours long and consist at minimum of a frontal EEG, a parietal EEG and an EMG trace sampled at 256 Hz. Several recordings additionally contain LFP traces and/or unsorted multi-unit activity.

The data comprises four groups:

- a pilot data set,

- a test data set,

- a sleep deprivation data set, and

- an optogenetic stimulation data set.

The recordings in the pilot data set and the test data set have been annotated by 4-10 experienced sleep researchers from the Vyazovskiy group at the University of Oxford. They are ideally suited for benchmarking of automated methods for polysomnography. The recordings in the sleep deprivation data set and the optogenetic stimulation data set exhibit characteristics that are distinct from corresponding baseline recordings and are thus useful to test the resilience of automated methods to experimental manipulations. Sleep deprivation increases the amplitude of slow-wave activity during NREM and thus changes the spectral features used by many automated methods. The manipulation in the optogenetic stimulation data set increased the number of times the animal was (briefly) awake during sleep, resulting in increases in the transition probabilities from NREM or REM to the awake state compared to baseline recordings. Methods that leverage expectations of transition probabilities in their predictions would be expected to be sensitive to these changes.

This benchmark data set was created during the development of Somnotate, an automated vigilance state classifier (available at https://github.com/paulbrodersen/somnotate). The data collection is hence described in the following publications:

Brodersen et al. Somnotate: A probabilistic sleep stage classifier for studying vigilance state transitions. PLoS Comput Biol. 2024. DOI: 10.1371/journal.pcbi.1011793

Please consider citing this publication if you use Somnotate or this data set in your academic work.