During non-rapid eye movement (NREM) sleep, neurons in the cortex alternate between periods of spiking (UP state) and periods of inactivity (DOWN state). In the current post, I have described how to detect UP and DOWN states within NREM epochs. 

Acknowledgments Dr. Justin Shobe for providing me with the test dataset; detect_peaks library by Marcos Duarte. NOTE, This script is just a starting script and there are other ways of going about solving the problem (Markov models).

The script is available HERE.

Algorithm

1. Load the raw LFP data along with the sampling rate (Fs)
2. Bandpass filter the data in 0.1-12 Hz.
3. Compute delta/theta power ratio (P_relative) to detect NREM and REM epochs. delta: 0-4Hz, theta: 7-12Hz
4. Compute epochs where the power ratio >= mean + 0.25*std of the P_relative and mark them as NREM epochs. The relative power threshold should be modified as per the data (high white line in 2nd-row image, 1st row: green background=NREM, yellow=REM).
5. Load the spiking data and calculate the sum of spikes from all the cells in time bins of 25ms. Time bin-width should be tested for individual datasets.
6. Set an up-state detection threshold = 5 spikes and falloff threshold = 1 spike`.
7. Use the up-state detection threshold and falloff threshold to calculate the up-state and down-state start and end timestamps.
8. Remove putative up-states and down-state which do not lie between the duration threshold (‘magenta’ color in 3rd and 4th row marks the up-state epoch).

I hope this analysis will be a great starting point for everyone who wants to deep dive into the sleep replay literature.