Exchangability blocks for nbse connectomestats

connectomics

#1

Hi all,

Following similar discussions with repeated measures analysis for FBA here and here, I was wondering if there was any scope for this development of longitudinal designs within connectomestats?

I can obviously do so with NBS1 in matlab, but I would like to take advantage of the nbse approach.

Oh, and admittedly I’m a newbie to connectomestats

For input of the connectivity matrices, does it require 1) the original asymmetric matrices from tck2connectome, or 2) can I use symmetric ones - say if I have thresholded (gulp) my networks?

Best,

Alistair


#2

Hi Alistair,

I’m endeavouring to make the statistical inference code - both current and future - as homogeneous as possible across different statistical inference commands. So yes, once I have sign-flipping working, it’ll appear simultaneously in fixelcfestats, mrclusterstats and connectomestats. For the use of exchangability blocks, we’ll probably stick to the current solution, which is simply to pre-generate the permutations using e.g. PALM and feed them in directly using the -permutations option.

Regarding the matrices, I need to slightly disentangle a couple of concepts:

  • Currently, connectomestats only operates on symmetric matrices; that is, the directionality of connections is not taken into account during the statistical testing, enhancement, or inference. Using NBSE on asymmetric matrices would require an altered enhancement algorithm; the design for which there is not a unique solution.

  • Although connectomestats operates on symmetric matrices, it is somewhat flexible in terms of the format of the matrices you provide. It will accept upper triangular, lower triangular, or full symmetric matrices. However, it will report an error if one of the input matrices is completely filled but contains inequal symmetric elements.

Cheers
Rob


#3

Good to hear @rsmith!

Ok, sorry I was loosely using the terms “asymmetric” and “symmetric”.

I was rather referring to the structure of the adjacency matrices. Say when calculating graph-properties in matlab, the networks are required to be symmetrized so that the upper triangle = lower triangle - i.e. [i,j] = [j,i]

But, the connectomes constructed with MRtrix only have the upper triangle? My question more specifically, can I input matrices with both a lower and upper triangle entries?

Cheers,

Alistair


#4

But, the connectomes constructed with MRtrix only have the upper triangle? My question more specifically, can I input matrices with both a lower and upper triangle entries?

Although connectomestats operates on symmetric matrices, it is somewhat flexible in terms of the format of the matrices you provide. It will accept upper triangular, lower triangular, or full symmetric matrices. However, it will report an error if one of the input matrices is completely filled but contains inequal symmetric elements.

:+1: