Phase amplitude coupling

[Noam Nitzan]

Dear Mike,

I have a question regarding the phase-amplitude coupling and in particular the exploratory aspect covered in chapter 30 of the book.

In the MATLAB code, you use the variables cfc_time_window and cfc_time_window_idx to define the window in data around which you run the analysis.

For figure 3.7 those variables are defined as:

cfc_time_window     = cfc_numcycles*(1000/freq4phase);
cfc_time_window_idx = round(cfc_time_window/(1000/EEG.srate));

In figure 30.8A you select one frequency for the power, and loop over phases. However, in the example code, the cfc_time_window variable doesn't change with every loop but remains set to the values defined in a previous cell.

Is this intended, or simply some oversight?

shouldn't cfc_time_window  be defined inside the for loop as:

cfc_time_window    = cfc_numcycles*(1000/phase_freqs(fi)) ? 

Adding this line to the loop obviously results in very different results (peak at 13 Hz no longer visible)..

Another, more general, question I had about the PAC is about an alternative method I saw in a couple of publications (e.g. Schomburg et al., Neuron 2014) which involves binning the phase time-series into phase intervals and computing the mean wavelet power for each of them.

A PAC value can then be computed by measuring the divergence of the observed amplitude distribution from the uniform distribution.

This method is quite different computationally from the classical one you present in the book (phase-amplitude product vs. mean amplitude for each phase), but intuitively should lead to similar conclusions.

Could you express your opinion about when to use each of these measures and what are the pros and cons for each of them? 

Many thanks in advance and hope you're having fun at the RSS!

Noam

[Mike X Cohen]

Hi Noam. See below. 

On Thu, Aug 8, 2019 at 11:39 AM Noam Nitzan <nnitz...@gmail.com> wrote:

Dear Mike,

I have a question regarding the phase-amplitude coupling and in particular the exploratory aspect covered in chapter 30 of the book.

In the MATLAB code, you use the variables cfc_time_window and cfc_time_window_idx to define the window in data around which you run the analysis.

For figure 3.7 those variables are defined as:

cfc_time_window     = cfc_numcycles*(1000/freq4phase);
cfc_time_window_idx = round(cfc_time_window/(1000/EEG.srate));

In figure 30.8A you select one frequency for the power, and loop over phases. However, in the example code, the cfc_time_window variable doesn't change with every loop but remains set to the values defined in a previous cell.

Is this intended, or simply some oversight?

shouldn't cfc_time_window  be defined inside the for loop as:

cfc_time_window    = cfc_numcycles*(1000/phase_freqs(fi)) ? 

Adding this line to the loop obviously results in very different results (peak at 13 Hz no longer visible)..

Good observation. I couldn't say now whether that was intentional or oversight. Although I generally would recommend having the time window be informed by the frequency for such a large range (2 to 20 Hz in this case). So perhaps it was the latter. The two methods produce similar-looking PAC up to 10 Hz, and then diverge. It's difficult to say whether that's due to differences in neural or cognitive factors, or to a change in SNR from the number of cycles.

 

Another, more general, question I had about the PAC is about an alternative method I saw in a couple of publications (e.g. Schomburg et al., Neuron 2014) which involves binning the phase time-series into phase intervals and computing the mean wavelet power for each of them.

A PAC value can then be computed by measuring the divergence of the observed amplitude distribution from the uniform distribution.

This method is quite different computationally from the classical one you present in the book (phase-amplitude product vs. mean amplitude for each phase), but intuitively should lead to similar conclusions.

Could you express your opinion about when to use each of these measures and what are the pros and cons for each of them? 

There are many measures of CFC, and new ones appearing each year. I think one could write an entire book just on this topic ;) 

Measuring KL distance of the distribution of amplitude over phase bins relative to a uniform (H0) distribution is also a good quantification method, and I think was first introduced by Adriano Tort. I'm not sure if there are really good ways of knowing which method to apply when. Honestly, my suspicion is that CFC in general is a relatively small phenomenon in most cases, and so some methods will happen to work better in some datasets and other methods will happen to work better in other datasets. Perhaps in the future there will be some theoretical motivation, but I'm not sure we're there yet.

 

Many thanks in advance and hope you're having fun at the RSS!

My week-long course at RSS is going great! I told them that they are the best group of students, but between you and me -- your class was the best. Just don't tell anyone else, lest they get jealous ;)

 

Noam

--
You received this message because you are subscribed to the Google Groups "AnalyzingNeuralTimeSeriesData" group.
To unsubscribe from this group and stop receiving emails from it, send an email to analyzingneuraltimes...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/analyzingneuraltimeseriesdata/8deca983-5cb7-4ac4-b5db-4ab4a0f801a4%40googlegroups.com.

--

Mike X Cohen, PhD
Fresh look: mikexcohen.com