Pactools: epoched data

💬 Support & Discussions
1

Hi,

I’m trying to compute theta-gamma PAC using pactools.

I compute Comodulogram and use the method .fit().
However, since my data is already epoched and concatenated across participants, I do not use raw_to_mask() function as suggested by the example.

Instead, I filter my data in the frequencies I would like to compute my PAC on:

low_sig = filter_data(data, sfreq, theta_band[0], theta_band[1], method="fir")
high_sig = filter_data(data, sfreq, gamma_band[0], gamma_band[1], method="fir")

After that, I use them in .fit() method:

estimator = Comodulogram(fs=sf,
                        low_fq_range=theta_range,
                        high_fq_range=gamma_range,
                        method='tort',
                        progress_bar=True)
# compute the comodulogram
estimator.fit(low_sig, high_sig)

The results I’m getting look very reasonable (at least I see what I expect to be seeing), but I’m still doubting on whether I’m doing the right thing by passing the filtered data to .fit().

I would appreciate any comments.
Thank you!

Best,
Kat

2

Hi,

Starting from a different point of view, if you were to use method="hagihira" (bicoherence) the fit call would return an error saying that it requires the same signal. Passing in only a low_sig will still return a comodulogram along the frequencies specified by low_fq_range and high_fq_range, which is actually the case for any method as shown in this example.

You can also check in the part of the documentation about STANDARD_PAC_METRICS that low_sig and high_sig are independently filtered with multiple (FIR) band-pass filters.

From that I assume that certain methods (BICOHERENCE_PAC_METRICS) are designed to work only within the same epochs (low_sig) and others (most of them) either within or across multiple epochs (low_sig & high_sig corresponding to DRIVER & CARRIER signals respectively) and that, since the filtering is handled internally, there is really no need for you to apply it beforehand.

Alternatively, I would assume that if you apply some filtering as you do it should be fairly broadband, because else there would be nothing to be found with PAC there. So, as long as your gamma band is not very narrow, I guess it’s still ok.

I think it would be interesting to try both approaches and try to figure out potential differences. If you do, I would be very interested to see the similarities and differences.

I hope this helps a bit.

Cheers,

1 Like
3

Hi,

That was indeed very helpful! Thanks :slight_smile:

1 Like
4

Cool!

If you figure out any interesting differences when filtering your data beforehand or not, please do share a pic!

Cheers,

5

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.