3dTfitter and gPPI modeling

AFNI Message Board
1

Hello AFNI team,

I am following the instructions here to complete a gPPI analysis: https://afni.nimh.nih.gov/CD-CorrAna

In Step 3 it recommends plotting the deconvolved seed timeseries to see if it looks reasonable when comparing to the stimulus presentation in the experiment. The data I am analyzing is a fast, event-related design, therefore I am wondering how to validate this step? I have used 1dplot to look at the timeseries before and after deconvolution, although I am not sure how to evaluate these plots.

In addition, for step 6 when running the regression, do I include all of my conditions of interest into one 3dDeconvolve step? For example, my code is pasted below with four conditions of interest (hour, day, week, month). Or should I run 3dDeconvolve four separate times instead, one for each condition?

Thank you,
Catherine

3dDeconvolve -input allruns+orig
-polort 3
-num_stimts 22
-mask /usr/local/mridata/Consolidation_2019/atlases/Kirby/dil+orig
-stim_file 1 /usr/local/mridata/Consolidation_2019/${i}/${i}.results.secondhalf.noblur/motion_short.1D’[0]’ -stim_base 1 -stim_label 1 roll
-stim_file 2 /usr/local/mridata/Consolidation_2019/${i}/${i}.results.secondhalf.noblur/motion_short.1D’[1]’ -stim_base 2 -stim_label 2 pitch
-stim_file 3 /usr/local/mridata/Consolidation_2019/${i}/${i}.results.secondhalf.noblur/motion_short.1D’[2]’ -stim_base 3 -stim_label 3 yaw
-stim_file 4 /usr/local/mridata/Consolidation_2019/${i}/${i}.results.secondhalf.noblur/motion_short.1D’[3]’ -stim_base 4 -stim_label 4 dS
-stim_file 5 /usr/local/mridata/Consolidation_2019/${i}/${i}.results.secondhalf.noblur/motion_short.1D’[4]’ -stim_base 5 -stim_label 5 dL
-stim_file 6 /usr/local/mridata/Consolidation_2019/${i}/${i}.results.secondhalf.noblur/motion_short.1D’[5]’ -stim_base 6 -stim_label 6 dP
-stim_file 7 /usr/local/mridata/Consolidation_2019/${i}/${i}.results.secondhalf.noblur/motion_deriv_short.1D’[0]’ -stim_base 7 -stim_label 7 rolldx
-stim_file 8 /usr/local/mridata/Consolidation_2019/${i}/${i}.results.secondhalf.noblur/motion_deriv_short.1D’[1]’ -stim_base 8 -stim_label 8 pitchdx
-stim_file 9 /usr/local/mridata/Consolidation_2019/${i}/${i}.results.secondhalf.noblur/motion_deriv_short.1D’[2]’ -stim_base 9 -stim_label 9 yawdx
-stim_file 10 /usr/local/mridata/Consolidation_2019/${i}/${i}.results.secondhalf.noblur/motion_deriv_short.1D’[3]’ -stim_base 10 -stim_label 10 dSdx
-stim_file 11 /usr/local/mridata/Consolidation_2019/${i}/${i}.results.secondhalf.noblur/motion_deriv_short.1D’[4]’ -stim_base 11 -stim_label 11 dLdx
-stim_file 12 /usr/local/mridata/Consolidation_2019/${i}/${i}.results.secondhalf.noblur/motion_deriv_short.1D’[5]’ -stim_base 12 -stim_label 12 dPdx
-stim_times 13 …/timing_files/${i}‘_all_foils_tf_chop5.txt’ ‘TENT(0,14,8)’ -stim_label 13 all_foils
-stim_times 14 …/timing_files/${i}‘_hour_targets_tf_chop5.txt’ ‘TENT(0,14,8)’ -stim_label 14 hour_targets
-stim_times 15 …/timing_files/${i}‘_day_targets_tf_chop5.txt’ ‘TENT(0,14,8)’ -stim_label 15 day_targets
-stim_times 16 …/timing_files/${i}‘_week_targets_tf_chop5.txt’ ‘TENT(0,14,8)’ -stim_label 16 week_targets
-stim_times 17 …/timing_files/${i}‘_month_targets_tf_chop5.txt’ ‘TENT(0,14,8)’ -stim_label 17 month_targets
-stim_file 18 AllRuns_HIPP_Seed.1D -stim_label 18 Seed
-stim_file 19 AllRuns_Interaction_timeseries_hour.1D -stim_label 19 PPIHour
-stim_file 20 AllRuns_Interaction_timeseries_day.1D -stim_label 20 PPIDay
-stim_file 21 AllRuns_Interaction_timeseries_week.1D -stim_label 21 PPIWeek
-stim_file 22 AllRuns_Interaction_timeseries_month.1D -stim_label 22 PPIMonth
-jobs 16
-rout -tout
-bucket fxnConnOutput

2

Hello AFNI team,

I was wondering if there was a solution to these issues?

Thank you,
Catherine

3

Hi Catherine,

With a fast event-related design, even the non-PPI regressors tend to look like noise, and certainly the PPI ones will. Also, if events are short, the PPI becomes more problematic, as it is very difficult to distinguish what happens only within the time of stimulus presentation when working with BOLD-convolved MRI data. In any case, plotting the PPI regressors will probably not fill you with confidence.

And yes, it is generally recommended to apply the gPPI regressors in a full model.

Note that there is a somewhat different but more complete example of PPI analysis here: AFNI_data6/FT_analysis/PPI. The cmd.ppi.2.make.regs script goes through creation of the gPPI regressors.

Again, working with short events makes this process even less stable than usual.

  • rick