FBA/stats questions


#1

Dear experts
I have a couple of questions regarding multi-tissue FBA pipeline which I would greatly appreciate if you kindly comment on.

  1. Should I use the -cutoff value of point 0.08 in tckgen (step 19) in case I increase the -fmls_peak_value to 0.08 in fod2fixel (step 12)? I guess the answer is “yes” but just want to be sure. :slight_smile:

  2. In a life span study (20-75 years of age), should I use intracranial volume (ICV) as a covariate of no interest when analyzing FD/logFC/FDC?

  3. Do these design (shortened version) and contrast look fine if I consider only younger and older groups and want to test whether there is any fixel showing an interaction group x performance?
    Design1:
    GR1 GR2 GR1per GR2per
    1 0 a1 0
    1 0 a2 0
    1 0 a3 0
    0 1 0 b1
    0 1 0 b2

where a1, a2,a3, b1, b2 are demeaned values “across both groups”. >> actually not sure if this is correct and when should be otherwise (i.e., within groups demeaning).
contrast 1:
[0 0 1 -1] >> add -neg to the fixelcfestats to compute [0 0 -1 1] as well.

And in case of adding ICV:
Design 2:
GR1 GR2 GR1per GR2per ICV
1 0 a1 0 c1
1 0 a2 0 c2
1 0 a3 0 c3
0 1 0 b1 c4
0 1 0 b2 c5
contrast 2:
[0 0 1 -1 0] >> add -neg to the fixelcfestats to compute [0 0 -1 1 0] as well.
Where a1, a2 , b1, b2, c1, c2,…,c5 are again demeaned values “across both groups”.

In the above situations, it would be nicer not to include a column of 1s. Right?

Cheers,
Hamed


#2

Hi Hamed,

Just to let you know (but you probably know by now): I’ve answered 1. and 2. via email, along with your other “internal” questions for your data specifically. The answers to both of those questions may vary based on application / research question / certain properties or qualities of the data. Especially question 1. can be tricky. General stuff here for reference (more specific answers in the email):

  1. It would seem to make sense to answer “yes” at first sight and in most scenarios, but that’s not necessarily the case. A "-cutoff < -fmls_peak_value" scenario can make good sense (and even often does in practice), where fixels are still connected up to a certain extent, even if a fixel “in between” doesn’t quite make the -fmls_peak_value threshold. Tractography has other constraints to make sure some things don’t run “out of control”. As long as you check the tckgen output on the template, and it looks good (no gross false positive connections, e.g. between hemispheres where they shouldn’t be, or gross false negative gaps in the WM), you’re good for that one. The -fmls_peak_value can be tweaked independently if needed, depending on your specific goal at that step.

  2. Entirely depends on what is in fact of interest (or not) to you. …but for a life span study, an effect of ICV is so trivial, that it almost certainly is not of interest. So for most life span studies, the answer is probably “yes”.

Cheers,
Thijs


#3

Hi Thijs

Thanks for the responses in both here and the email.
Could you please also comment on the design and the demeaning in the design?

Thanks again and cheers,
Hamed


#4

Oh, yeah, that certainly looks entirely fine; no worries there, I’d say! :+1: Demeaning across both groups seems entirely sensible. If I understand correctly what you’re referring to with “within groups demeaning”, that would be a strange thing to do here, since it would do away with the interaction you’re looking for, right?


#5

edit: “within groups demeaning” >> “demeaning within each group”


#6

I think we’re talking about the same thing then. :+1: (as in “using the individual mean of each group”)