Hi Experts!

In my first level GLM, I would like to include several parametric modulators like in this paper. I am interested in these parametric modulators and not the main regressors.

My data was preprocessed in fmriprep and I have been developing my firstlevel pipeline in nilearn. I have found this post on parametric modulation in nilearn VERY helpful, however, much to my disappointment, nilearn doesn’t seem to support multiple parametric modulators. Here is a discussion on this issue:Optional orthogonalization of regressors · Issue #2384 · nilearn/nilearn · GitHub, and in the same discussion **bthirion** mentioned that there is a way to by pass this by duplicating the event.

Does this mean that the events should be duplicated vertically along with the same onset time and duration, but with different modulation values? I suspect it is this way because I came across another discussion about duplicated events where @tsalo mentioned

“duplicate events should be treated as a valid substitute for amplitude modulation”

I assume he was referring to this?

However, upon trying it myself, I got a message:“Amplitudes of these events will be summed” which if I am not wrong was implemented sometime ago? This is not what I hope to achieve. I need to be able to separate the effect of each parametric modulator.

Alternatively, is it alright if I ran multiple GLMs for each modulators instead of 1 for all of them? My intuition is that this might not be statistically sound be the model is not “full” as some of the regressors are omitted?

If not I know SPM and FSL allows multiple parametric modulators, I can use nipype to implement that as well, but I can’t seem to find how to input the modulators into the model. Can anyone point me to a nipype tutorial on multiple parametric modulators?

Thank you!

Regards,

Vae

More information if it helps:

In my task-fMRI experiment, I am interested in the bayesian inference behaviours of different age groups (Young/Old) under probabilistic conditions (regressors: high_probability /low_probability). Thus in my first level analysis I am not particularly interested in response of the brain to the probabilistic conditions(i.e. the regressors) but the parametric modulation of the bayesian inference parameters (such as level of prediction errors etc) to the high_probability and low_probability regressors in the brain. I have several such bayesian inference parameters that I calculated from subject’s behavioural output collected in the same task-fMRI experiment.