Parameter Exploration in Jansen Rit

[Orestis Stylianou]

Hello, 

I have some healthy EEG signals and I try to simulate them. I was told that the best neural mass model for this is Jansen Rit. 

I am stuck in parameter exploration. I was told to start with global coupling (G). First, I started in the range of 0 to 1200 with increments of 10. I then proceeded with higher ranges, until I reached 200000, again with correlations less than 0.01. Right now, I am running a much broader parameter exploration from 200000 to 2000000000 with increments of 50. 

I am wondering if my strategy makes sense.  It seems to me that there is no expected range to explore. Am I missing something? Theoretically, I could spend months upon months searching for good values of G without any result... 

Thanks!

[Orestis Stylianou]

Edited: 

Hello, 

I have some healthy EEG signals and I try to simulate them. I was told that the best neural mass model for this is Jansen Rit. 

I am stuck in parameter exploration. I was told to start with global coupling (G). First, I started in the range of 0 to 1200 with increments of 10, the average correlation between the original and simulated signals was less than 0.01. I then proceeded with higher ranges, until I reached 200000, again with correlations less than 0.01. Right now, I am running a much broader parameter exploration from 200000 to 2000000000 with increments of 50. 

I am wondering if my strategy makes sense.  It seems to me that there is no expected range to explore. Am I missing something? Theoretically, I could spend months upon months searching for good values of G without any result... 

Thanks!

[NeuroLife]

Hi,

Depending on the structural connectivity that you have the value of the global coupling coefficient can be smaller. So, perhaps you might want to do a log10 scale search from say 1e-5 to 1e1. This might help find out if the required global coupling coefficient is of a smaller scale than what you've been searching for so far. Alternatively, you could look into how your structural connectivity matrix is scaled and if it's normalized. You could also plot the time series to see how the simulations behave. This might help you ensure that the issue is not just caused by the chosen parameters of local neural mass model used. There's a few other things you could look into but I think this might be a good start.

Hope this helps!

Dominic Boutet 

[Orestis Stylianou]

Hello,

Thanks for the input. I am trying to create a plug n play approach, so I am using the default TVB connectivity. I just started with high values of G because some papers used similar values. But I will try some smaller values too. I have also plotted some of the simulated time series. They do look like EEG, so I don't think I am in a damped oscillation or anything like that... 

Best, Orestis!

[WOODMAN Michael]

What is the connectome you’re using? You might want to compare the signals in terms of power spectra, not just the time series.

> On 24 Dec 2022, at 09:48, Orestis Stylianou <oest...@gmail.com> wrote:
>
> Hello,

[Orestis Stylianou]

Hello, 

Thank you for your answer! I am sorry for the late reply. The notification went to my SPAM folder. 
I am using the connectivity 76 zip. 

I wanted to compare the time series themselves because this is the ultimate goal of a simulation, to have almost perfect replicas of brain activity. It looks like this is a very ambitious project, especially for a beginner in TVB, so I might have to rely on my traditional methods like power spectra. 

Best, Orestis!