HMM states failed to capture different sleep stages

[nilsyang]

Hi Dr Vidaurre,

I am working on replicate the results of Stevner, et. al. Nature communications (2019) on our all-night EEG-fMRI sleep data. The setting I used is from 'run_HMMMAR_2.m' with '% Gaussian distribution with mean and full covariance".

However when I tried k (num of states) from 4 to 20. I found out the states were not differ from each other. and the stats.lambda (in manova1 compared with sleep stages) is close to 1 in almost all states. Below is an example of k=4, in which i plot each of 4 state against the sleep stages (0=n/a, 1=N4 , 2=N3, 3=N2, 4=N1, 5=rem, 6=wake). It looks like hmm states did not signifcant differ from each other.

I wonder if there is anything wrong in my settings that caused the problem.

[vidaurre]

Hi @nilsyang , it's difficult to know exactly what happens without having more details. In any case, it might be worth asking the first author, Angus Stevner, as he was the one in charge of the analysis. Feel free to drop us both an email.