Reduced WongWangExcInh low fits
Hannah Hacker
I started using TVB for my Masterthesis and want to run simulations with the Reduced Wong Wang ExcInh Model on HCP data. I used the structural and functional Connectivity matrices available here: https://search.kg.ebrains.eu/instances/f16e449d-86e1-408b-9487-aa9d72e39901 in the Desikan Killiany parcellation. Although I experimented with various parameter settings (mainly varying G, but also trying out variations for J_N, I_o, w_p, gamma_e and a_e seperately) I somehow get very low fits between the empirical FC and the simulated FC (<0.1). Because I realised that the SC matrices in the democases look quite different, especially that there are much more zeros, I tried to transform them in different ways (normalizing, MinMaxScaling) and got slightly better results (0.24), when using a RobustScaling approach, but still not what can be found in literature. The correlations between the SC and empirical FC look fine (0.26-0.45). For the simulated FC the correlations with the SC are very low (<0.1) for the simulations with the low fits and higher (0.42) for the transformed case.
So I wondered if there are any processing steps I need to perform on the SC, before they are suitable for using in TVB? Alternatively do you have any insights into potential factors, that could be the reason for the low fits?
Thanks in advance for your help!
Best,
Hannah
WOODMAN Michael
hi
What do you mean by robust scaling? What degree of fit are you trying to reproduce from the literature? Some of the very high fit results are simply a matter of averaging over subjects, while in more recent cases, it's enabled by varying certain parameters per region per subject.
cheers,
Marmaduke
Sent: Thursday, September 14, 2023 11:50:42 AM
To: TVB Users
Subject: [RESEAUX SOCIAUX] [TVB] Reduced WongWangExcInh low fits
You received this message because you are subscribed to the Google Groups "TVB Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tvb-users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/596205f2-5fb0-4216-ba40-d6bc27dac458n%40googlegroups.com.
Hannah Hacker
Hi,
tank you for your answer. I mean a way of rescaling that is robust to outliers by removing the median and scaling the data according to the quantile range. I saw correlations for example in the papers by Deco et al. that were a lot higher than .24, so I was expecting something of around 0.4. I also already used averaged SC matrices.
WOODMAN Michael
hi,
I see what you mean by robust. Sometimes we found it useful to normalize by the 99th percentile and clip values at 1 for similar reasons, as it maximized the eFC - sFC correlations. Nevertheless the modeling question is not always about the absolute correlation level, but what changes, groups or conditions induce a statistically significant improvement in correlation and how they're linked to model parameters.
cheers,
Marmaduke
To: TVB Users
Subject: [RESEAUX SOCIAUX] Re: [RESEAUX SOCIAUX] [TVB] Reduced WongWangExcInh low fits