Compute PAC using pactools

💬 Support & Discussions
1

Hi,

I’m trying to compute theta-gamma PAC using pactools.
I’d like to build a Comodulogram and fit it to the signal. In the example provided the signal is one-dimentional matrix of 10000 points. If we translate it into EEG field, it would be the signal from 1 channel, having 1 value for each of 10000 time points.

In my data, for each channel I have multipme epochs, which increases dimentionality of the data. Could you please suggest, what would be a proper way to reduce data dimentionality to compute PAC for all epochs on one channel simultaneously?

I know it is not within MNE infrastructure, but maybe someone have experience with that. I’ll appreciate your help!

Best,
Kat

2

see https://pactools.github.io/auto_examples/plot_mne_epoch_masking.html

you need to compute it on the raw object but you can pass a MaskIterator to take into account the epochs
structure.

Alex

2 Likes
3

In the Comodulagram class, the fit function definitely accepts matrices with dimensions (n_epochs, n_points): pactools.comodulogram — pactools 0.1 documentation.

To the extent I have tested the package with bicoherence and bispectrum measures, I am pretty sure that you would still get a result when using epochs as input.

Is there an advantage of running PAC on raw instead of epochs if you have the option to do so? What if you only have access to epochs and not raw data?

1 Like
4

Hi Alex,

Thanks for your reply!

The data I have at hand has epochs concatenated across multiple participants and I am not sure I can use the approach you’ve suggsted in this case (maybe I am wrong?).

Excuse me for the naive quesitons, but:

  1. how exactly does MaskIterator work when we pass events there? I’ve read the documentation, but I am still not getting how it a single comodulogram over epochs is eventually computed? my feeling is that averaging should be occuring at some point,
  2. is it possible to use the concatenated epochs file as a raw file and use masking on it? can I solve inconsistencies of the signal at the edges of the epochs in this case?
  3. would averaging all epochs across participants before running pac be a decent approach or would it rather blur the effects?

Please let me know, if I made my point clear. I am a bit confused at this stage :slight_smile:

Best,
Kat

5

Thanks a lot for your comment!

That is the thing that I have my epochs concatenated across participants and coming back to raw data structure would be challenging at this point. Hoping for some further clarification from Alexandre.