Creating trial-by-trial regressors

AFNI Message Board
1

Hello!

I’m trying to run some analyses that are conceptually similar to a PPI and parametric modulation. In a nutshell, I’m going to be using the trial-by-trial beta parameter from my anticipatory (cue) epoch as a parametric regressor for my feedback (FB) epoch. So the first step is to run individual level analyses to get the trial-by-trial coefficients for my cue epoch.

My task has an anticipatory (cue) epoch, then a feedback (FB) epoch. It has 72 trials and 3 runs. I need to run an individual level analysis that gives me a coefficient for the cue epoch for each of the 72 trials, so essentially i’d have analysis with 72 sub-briks, one for each cue onset of each trial.

Hopefully that makes sense.

Where I’m getting stuck is how best to model the regressors- I think what I want is 72 separate 1D files, each filled with the onset:duration of the cue epoch for that trial (1-72). However, I’m unsure what these should look like. For example, do I want for the first trial something like this?

9:2
*
*
This would account for the fact that this cue onset:duration trial is in run 1. Or do I just want something like this

9:2

which would simply be the onset:duration for the cue event for trial.

the next tricky thing is figuring out how to code this, but that I can figure out on my own. Unless you know of a nice tool that’s already programmed to split out your 1D files into trial-by-trial regressors!

I’m more than happy to provide any other information that would be helpful. Thank you so much!!

2

Hello,

Actually, that is what Bob calls IM (individual modulation). It would be sufficient to pass 3dDeconvolve the cue timing (across all trials, as originally done), but pass it with -stim_times_IM (rather than just -stim_times). The output will have one beta per event.

Give that a try.

  • rick
3

Oh wow that’s amazing! So much easier.

Just to clarify, I’ll have a cue.1D file with 3 rows (because I have 3 runs) which I’ll use the -stim_times_IM flag for

I’ll also have a bunch of regressors of no interest (feedback epoch onsets)

I THINK I’ve gotten it coded up correctly if this looks good? THANK YOU SO MUCH!

-regress_stim_times	   																		\
		    $stimdurmoddir/anticipation.1D													\
			$stimdurmoddir/Mean_feedback.1D													\
		    $stimdurmoddir/Mean_response.1D													\
		    $stimdurmoddir/Nice_feedback.1D													\
		    $stimdurmoddir/Nice_response.1D													\
		    $stimdurmoddir/UnpMean_feedback.1D												\
		    $stimdurmoddir/UnpMean_response.1D												\
		    $stimdurmoddir/UnpNice_feedback.1D												\
		    $stimdurmoddir/UnpNice_response.1D												\
-regress_stim_labels																		\
			Anticipation																	\
		    FB.Mean																			\
		    Resp.Mean																		\
		    FB.Nice																			\
		    Resp.Nice																		\
		    FB.UnpMean																		\
		    Resp.UnpMean																	\
		    FB.UnpNice																		\
		    Resp.UnpNice																	\
-regress_stim_types 	   																	\
							'IM'															\
							'AM1'															\
							'AM1'															\
							'AM1'															\
							'AM1'															\
							'AM1'															\
							'AM1'															\
							'AM1'															\
							'AM1'															\
-regress_basis_multi 																		\
						'dmBLOCK(0)'														\
4

Yes, you can pass a bunch of normal -stim_times files, as well as even a single -stim_times_IM.

The place to verify what is happening is in the regression matrix (which might be hard to plot, say). But even something like this can clarify the details:

1d_tool.py -show_group_labels -infile X.xmat.1D

You should still see the 3 runs of events (for the IM case) when plotting the X-matrix, but they will be spread across the regressors.

  • rick
5

Thank you, Rick!

Going off of this other help now that I have the “individual modulation” language… https://afni.nimh.nih.gov/afni/community/board/read.php?1,156358,156359#msg-156359

I THINK I’ve gotten it coded up correctly if this looks good? THANK YOU SO MUCH!

-regress_stim_times
$stimdurmoddir/anticipation.1D
$stimdurmoddir/Mean_feedback.1D
$stimdurmoddir/Mean_response.1D
$stimdurmoddir/Nice_feedback.1D
$stimdurmoddir/Nice_response.1D
$stimdurmoddir/UnpMean_feedback.1D
$stimdurmoddir/UnpMean_response.1D
$stimdurmoddir/UnpNice_feedback.1D
$stimdurmoddir/UnpNice_response.1D
-regress_stim_labels
Anticipation
FB.Mean
Resp.Mean
FB.Nice
Resp.Nice
FB.UnpMean
Resp.UnpMean
FB.UnpNice
Resp.UnpNice
-regress_stim_types
‘IM’
‘AM1’
‘AM1’
‘AM1’
‘AM1’
‘AM1’
‘AM1’
‘AM1’
‘AM1’
-regress_basis_multi
‘dmBLOCK(0)’ \

6

Hmmm, so the IM class is being modeled with dmBLOCK? I guess that should be okay, though I have never come across it, an extra complication…

  • rick
7

I can take the duration modulation out of it if that’s best, they’re all the same duration so it shouldn’t be an issue- that’s just how we typically have modeled our stim files in the past.

8

Well, I would not say “best”, it is just more complicated to have varying duration. But if they are all of the same duration anyway (across events and subjects), then it does not even matter (aside from being a little confusing if you publish the afni_proc.py command).

If they are the same duration D, I would probably go with BLOCK(D,1) for clarity, but that is up to you.

  • rick
9

Thank you so much, Rick. I really appreciate all your help! I’ll reach out after I’ve run some individual levels and take a look at the output if anything looks odd.

10

Sounds good, thanks!

  • rick