Thank you for your help, now I have this as my updated AFNI processing pipeline:
I had to include a GOFORIT 1 because I got the following error from the adding the highpass.1D file. Since it is just defining the baseline model, collinearity between the regressors isn't that much of a problem, right?
Error Message:
++ ==========================================================
*+ WARNING: -------------------------------------------------
*+ WARNING: Problems with the X matrix columns, listed below:
*+ WARNING: !! * Column 11 [ortvec[0]#0] is all zeros
*+ WARNING: -------------------------------------------------
++ ----- Signal+Baseline matrix condition [X] (315x24): 7.06122 ++ VERY GOOD ++
*+ WARNING: !! in Signal+Baseline matrix:
* Largest singular value=2.3688
* 1 singular value is less than cutoff=2.3688e-07
* Implies strong collinearity in the matrix columns!
++ Signal+Baseline matrix singular values:
1.49012e-08 0.0475083 0.0840822 0.173134 0.266421
0.322178 0.447562 0.61145 0.680342 0.766082
0.772279 1 1 1 1.00064
1.00153 1.00236 1.0232 1.06652 1.11331
1.17726 1.19201 1.6476 2.3688
++ ----- Signal-only matrix condition [X] (315x10): 4.69282 ++ VERY GOOD ++
++ ----- Baseline-only matrix condition [X] (315x14): 1 ++ VERY GOOD ++
*+ WARNING: !! in Baseline-only matrix:
* Largest singular value=1
* 1 singular value is less than cutoff=1e-07
* Implies strong collinearity in the matrix columns!
++ Baseline-only matrix singular values:
0 1 1 1 1
1 1 1 1 1
1 1 1 1
++ ----- stim_base-only matrix condition [X] (315x13): 1 ++ VERY GOOD ++
*+ WARNING: !! in stim_base-only matrix:
* Largest singular value=1
* 1 singular value is less than cutoff=1e-07
* Implies strong collinearity in the matrix columns!
++ stim_base-only matrix singular values:
0 1 1 1 1
1 1 1 1 1
1 1 1
++ ----- polort-only matrix condition [X] (315x1): 1 ++ VERY GOOD ++
++ 3dDeconvolve exits: -x1D_stop option was invoked
++ 3dREMLfit: AFNI version=AFNI_2011_12_21_1014 (Sep 30 2014) [64-bit]
++ Authored by: RWCox
++ Number of voxels in mask = 29484
++ GOFORIT ==> Matrix de-singularization is engaged!
++ Number of OpenMP threads = 12
** ERROR: matrix column #11 is all zero!?
*+ WARNING: You said to GOFORIT, so here we GO!
+ Note that a bunch of further WARNINGs will be generated below.
++ Denominator DOF increased from 291 to 292 to allow for all zero columns
++ ----- matrix condition (315x24): 7.06122 ++ VERY GOOD ++
*+ WARNING: !! in matrix:
* Largest singular value=2.3688
* 1 singular value is less than cutoff=2.3688e-07
* Implies strong collinearity in the matrix columns!
++ matrix singular values:
0 0.0475083 0.0840822 0.173134 0.266421
0.322178 0.447562 0.61145 0.680342 0.766082
0.772279 1 1 1 1.00064
1.00153 1.00236 1.0232 1.06652 1.11331
1.17726 1.19201 1.6476 2.3688
*+ WARNING: -GOFORIT ==> Charging ahead into battle!
+ ==> Check results carefully!
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
preprocessing steps::
motion correction::: 3dvolreg (AFNI)
spatial smoothing::: susan (FSL)
1dBport -band 0 0.01 -input ${FLANKER_NIFTI} > highpass.1D #NEW LINE
3dDeconvolve -input ${FLANKER_NIFTI} \
-nfirst 0 \
-polort 0 \
-num_stimts 10 \
-mask ${outDir}/mask/*_mask.nii.gz \
-stim_times 1 ${timing_array[0]} 'GAM(6.0024,0.9996)' -stim_label 1 con \
-stim_times 2 ${timing_array[1]} 'GAM(6.0024,0.9996)' -stim_label 2 errors \
-stim_times 3 ${timing_array[2]} 'GAM(6.0024,0.9996)' -stim_label 3 inc \
-stim_times 4 ${timing_array[3]} 'GAM(6.0024,0.9996)' -stim_label 4 neu \
-stim_file 5 ${motion_file}[0] -stim_label 5 roll \
-stim_file 6 ${motion_file}[1] -stim_label 6 pitch \
-stim_file 7 ${motion_file}[2] -stim_label 7 yaw \
-stim_file 8 ${motion_file}[3] -stim_label 8 I_S \
-stim_file 9 ${motion_file}[4] -stim_label 9 R_L \
-stim_file 10 ${motion_file}[5] -stim_label 10 A_P \
-num_glt 8 \
-glt_label 1 con_ave -gltsym 'SYM: con' \
-glt_label 2 errors_ave -gltsym 'SYM: errors' \
-glt_label 3 inc_ave -gltsym 'SYM: inc' \
-glt_label 4 neu_ave -gltsym 'SYM: neu' \
-glt_label 5 con-neu -gltsym 'SYM: +con -neu' \
-glt_label 6 inc-neu -gltsym 'SYM: +inc -neu' \
-glt_label 7 con-inc -gltsym 'SYM: +con -inc' \
-glt_label 8 inc-con -gltsym 'SYM: +inc -con' \
-ortvec highpass.1D \ #NEW LINE
-tout -fout -bucket sub${subNum}_bucket -xjpeg sub${subNum}_glm_matrix.jpg -x1D_stop &&\
3dREMLfit -matrix sub${subNum}_bucket.xmat.1D \
-GOFORIT 1 \ #NEW LINE
-input ${FLANKER_NIFTI} \
-mask ${outDir}/mask/*_mask.nii.gz \
-fout -tout -Rbuck sub${subNum}_bucket_REML -Rvar sub${subNum}_bucket_REMLvar -verb
The results still appear to be much different from FSL's output and fewer voxels were considered significant in the new AFNI command (since more was included in the baseline model reducing degrees of freedom, maybe?), but the results were less "noisy" looking. Perhaps at the single subject level, it is typical for AFNI to give different results from FSL? At least for my dataset.
Thanks again!
James
