This question is about the COBRE dataset derivative available here on figshare: https://figshare.com/articles/dataset/Cobre_for_machine_learning/1450804/2
In the dataset there are 1D arrays clearly representing halves of the functional connectivity matrices extracted at different resolutions.
My problem lies in the fact that I cannot reconstruct the 2D matrix from its 1D array half as I do not know which elements of the 1D array represent which elements of the matrix.
Are there any conventions in the neuroscience community (or broader) for how to reconstruct a symmetrical matrix which is given as one of its halves in a 1D array?
I tried contacting the authors of the dataset, but did not get a response.
Thank you in advance!
I would take a wild guess that the 1D array is obtained using the triu(matrix_2D) function; get values in the upper triangle. Some network-based statistics packages assume that 1D arrays represent the values in the upper triangle.
In anycase, i have preprocessed (fMRIprep) the COBRE dataset, and would be happy to share my FC matrices if you’re interested.
Thank you for your reply!
I’d gladly accept your FC matrices.
One more question, I have a preprocessed fMRI dataset with a confounds array which I’ve tried to use for confound regression through NiLearn when building FC matrices. But, when I plot them, almost all of the elements in the matrix are above 0.5, many close to 1 even. I believe this tells me I am doing something wrong, as all the matrices in the tutorials I’ve seen show most of the matrix elements as being close to zero after confounds regression.
Even if you messed up the confounds regression, i dont think you will end up with FC matrices where “almost all of the elements in the matrix are above 0.5”. I suspect your atlas parcellation file is not well-aligned to whatever standard space you using.
I’ll pm you regarding my COBRE FC matrices