Hello,
This is probably a quick question. It seems to take a long time to generate monte carlo simulation data in mri_glmfit. Is it a bad idea to break the iterations into however many available seychelles nodes I can use? Is that equivalent to running the total number of iterations serially?
Thanks, Rob
fpolli@nmr.mgh.harvard.edu wrote:
hi rob, see below for the answers to some of your questions.
Hi Dara and Frida,
The analysis is running. There is one part of Frida's example that I will need to emulate once this is done, but if what I have is correct, then the part that takes the most time will be done when I come in to work on Thursday 7/5.
Frida, sometime when you have a moment free, could you look over the commands I'm running and tell me if there is anything glaringly (or subtly) wrong? I have tried my best to run it in a way that is consistent with how you did it, but I think there could be things wrong, because I dont 100% understand everything about what these commands are doing, but getting there. There's nothing in these commands that looks like nonsense to me, but for example, I don't know if the smoothing value of 13.488420 is the same in our case.
to find this parameter, you want to open a file that's called mri_glmfit_log or something like that. it is created in the third step. however, it usually gives you a value b/w 12.5 and 14.5 and using one vs.the other does not create a huge diffrence. however, try to find the file i'm talking about, open it up, and you'll see.
I'm using a threshold of .01, and I
don't know if that would cause any parameters other than the threshold to need to change.
you should use whatever parameter you want your data displayed at. no that doesn't change anything else.
I also don't know why simulation data is generated
twice and then both input into the contrast,
only b/c it takes a long time to generate 10K simulations, so we broke it up into 2 x 5K. you can also use 1 x 10 k. but I'm following that
aspect of your example verbatim at least for now, in the commands that are now running. Also a default glmfit setting is to use dods (diff order diff slope).
dods is DEF. the one you want to use. it stands for different offset different slope - you don't want to assume that both groups have the same slope to their data.
Is that what we should be using? There is also a
doss (same slope) option. Ally, she is one of the RAs who I talked with yesterday about this, mentioned that dods is more unpredictable than doss.
Thanks for your help, Rob
good luck, f
#!/usr/local/bin/bash
################## # levy 7/3/07 # ##################
# Katy ran intergroupavg-sess on 9/20/05 # Last week we used FDR to correct for multiple comparisons on the results of the above contrast between groups, # but the resulting threshold was too conservative. # So now I am running the analysis with a cluster-based threshold using a Monte Carlo simulation # (simulate a null hypothesis with randomnes, and use it to generate a meaningful threshold to judge the results with) # In terms of the freesurfer implementation, this is a different process and does not build on the results of previous commands.
# so we start from the beginning # note: rerunning these 2 lines probably unnecessary, but mris_preproc was having problems I thought were due to faulty stxgrinder output # This problem led Doug to discover bugs in mris_preproc and also mkcontrast-sess, which he then fixed # As a result, I re-ran the following 2 commands, which didn't take very long anyhow.
mkcontrast-sess -analysis EMerror -contrast ASvfix_6 -a 2 -c 0 -setwdelay -sumdelays stxgrinder-sess -a EMerror -c ASvfix_6 -sf Subject_Lists/bw-group-error-5
# resample onto average7 # ( use version from doug's home directory with bugs corrected... )
/space/greve/1/users/greve/freesurfer/bin/mris_preproc --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --target average7 --hemi lh --sf Subject_Lists/bw-group-error-5 --a EMerror --c ASvfix_6 --out lh.ASvfix_8s_BwGroupError5.fsfast.mgh | tee -a logs/mris_preproc_BwGroupError5_ASvfix_8s_lh.log /space/greve/1/users/greve/freesurfer/bin/mris_preproc --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --target average7 --hemi rh --sf Subject_Lists/bw-group-error-5 --a EMerror --c ASvfix_6 --out rh.ASvfix_8s_BwGroupError5.fsfast.mgh | tee -a logs/mris_preproc_BwGroupError5_ASvfix_8s_rh.log
mkdir lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/ mkdir rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/
# lh mri_glmfit --really-use-average7 --y lh.ASvfix_8s_BwGroupError5.fsfast.mgh --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --glmdir lh.ASvfix_8s_BwGroupError5.fsfast.glmdir --surf average7 lh --C fsgd_and_contrast_files/BwGroupError5.mat --sim mc-full 5000 2 BwGroupError5_ASvfix_8s_lhMCSIM1 --fwhm 13.488420 | tee -a logs/glmfit_lh_MCSIM1.log mri_glmfit --really-use-average7 --y lh.ASvfix_8s_BwGroupError5.fsfast.mgh --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --glmdir lh.ASvfix_8s_BwGroupError5.fsfast.glmdir --surf average7 lh --C fsgd_and_contrast_files/BwGroupError5.mat --sim mc-full 5000 2 BwGroupError5_ASvfix_8s_lhMCSIM2 --fwhm 13.488420 | tee -a logs/glmfit_lh_MCSIM2.log mri_surfcluster --src lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.mgh --csd BwGroupError5_ASvfix_8s_lhMCSIM1-BwGroupError5.csd --csd BwGroupError5_ASvfix_8s_lhMCSIM2-BwGroupError5.csd --sum lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.clusterMCSIM.sum --ocp lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.clusterMCSIM.mgh --thmax infinity | tee -a logs/surfcluster_lh_BwGroupError5_AsvFix_mcsim.log # rh mri_glmfit --really-use-average7 --y rh.ASvfix_8s_BwGroupError5.fsfast.mgh --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --glmdir rh.ASvfix_8s_BwGroupError5.fsfast.glmdir --surf average7 rh --C fsgd_and_contrast_files/BwGroupError5.mat --sim mc-full 5000 2 BwGroupError5_ASvfix_8s_rhMCSIM1 --fwhm 13.488420 | tee -a logs/glmfit_rh_MCSIM1.log mri_glmfit --really-use-average7 --y rh.ASvfix_8s_BwGroupError5.fsfast.mgh --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --glmdir rh.ASvfix_8s_BwGroupError5.fsfast.glmdir --surf average7 rh --C fsgd_and_contrast_files/BwGroupError5.mat --sim mc-full 5000 2 BwGroupError5_ASvfix_8s_rhMCSIM2 --fwhm 13.488420 | tee -a logs/glmfit_rh_MCSIM2.log mri_surfcluster --src rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.mgh --csd BwGroupError5_ASvfix_8s_rhMCSIM1-BwGroupError5.csd --csd BwGroupError5_ASvfix_8s_rhMCSIM2-BwGroupError5.csd --sum rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.clusterMCSIM.sum --ocp rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.clusterMCSIM.mgh --thmax infinity | tee -a logs/surfcluster_rh_BwGroupError5_AsvFix_mcsim.log
# getting the correct map to display
# ( not yet done ) # # here is Frida's example # # mri_surfcluster --src lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.mgh --o lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_o.mgh --hemi lh --srcsubj average7 --thmin 1.3 # mri_surfcluster --src lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.mgh --csd BwGroup3b_ASevASc_6s_lh_MCSIM1-BwGroupError3b_NvSZ_contrast.csd --csd BwGroup3b_ASevASc_6s_lh_MCSIM2-BwGroupError3b_NvSZ_contrast.csd --thmax infinity --cwsig lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_cwsig.mgh
# mri_mask -T 1.3 lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_o.mgh lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_cwsig.mgh lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_ocwsig.mgh
--
Robert P. Levy, B.A. Research Assistant, Manoach Lab Massachusetts General Hospital Charlestown Navy Yard 149 13th St., Room 2656 Charlestown, MA 02129 email: levy@nmr.mgh.harvard.edu phone: 617-726-1908 fax: 617-726-4078
mri_glmfit was programmed to be able to break things up for easy parallel implementation. I don't know if the other users on seychelles will like it though:).
On Thu, 5 Jul 2007, Robert Levy wrote:
Hello,
This is probably a quick question. It seems to take a long time to generate monte carlo simulation data in mri_glmfit. Is it a bad idea to break the iterations into however many available seychelles nodes I can use? Is that equivalent to running the total number of iterations serially? Thanks, Rob
fpolli@nmr.mgh.harvard.edu wrote:
hi rob, see below for the answers to some of your questions.
Hi Dara and Frida,
The analysis is running. There is one part of Frida's example that I will need to emulate once this is done, but if what I have is correct, then the part that takes the most time will be done when I come in to work on Thursday 7/5.
Frida, sometime when you have a moment free, could you look over the commands I'm running and tell me if there is anything glaringly (or subtly) wrong? I have tried my best to run it in a way that is consistent with how you did it, but I think there could be things wrong, because I dont 100% understand everything about what these commands are doing, but getting there. There's nothing in these commands that looks like nonsense to me, but for example, I don't know if the smoothing value of 13.488420 is the same in our case.
to find this parameter, you want to open a file that's called mri_glmfit_log or something like that. it is created in the third step. however, it usually gives you a value b/w 12.5 and 14.5 and using one vs.the other does not create a huge diffrence. however, try to find the file i'm talking about, open it up, and you'll see.
I'm using a threshold of .01, and I
don't know if that would cause any parameters other than the threshold to need to change.
you should use whatever parameter you want your data displayed at. no that doesn't change anything else.
I also don't know why simulation data is generated
twice and then both input into the contrast,
only b/c it takes a long time to generate 10K simulations, so we broke it up into 2 x 5K. you can also use 1 x 10 k. but I'm following that
aspect of your example verbatim at least for now, in the commands that are now running. Also a default glmfit setting is to use dods (diff order diff slope).
dods is DEF. the one you want to use. it stands for different offset different slope - you don't want to assume that both groups have the same slope to their data.
Is that what we should be using? There is also a
doss (same slope) option. Ally, she is one of the RAs who I talked with yesterday about this, mentioned that dods is more unpredictable than doss.
Thanks for your help, Rob
good luck, f
#!/usr/local/bin/bash
################## # levy 7/3/07 # ##################
# Katy ran intergroupavg-sess on 9/20/05 # Last week we used FDR to correct for multiple comparisons on the results of the above contrast between groups, # but the resulting threshold was too conservative. # So now I am running the analysis with a cluster-based threshold using a Monte Carlo simulation # (simulate a null hypothesis with randomnes, and use it to generate a meaningful threshold to judge the results with) # In terms of the freesurfer implementation, this is a different process and does not build on the results of previous commands.
# so we start from the beginning # note: rerunning these 2 lines probably unnecessary, but mris_preproc was having problems I thought were due to faulty stxgrinder output # This problem led Doug to discover bugs in mris_preproc and also mkcontrast-sess, which he then fixed # As a result, I re-ran the following 2 commands, which didn't take very long anyhow.
mkcontrast-sess -analysis EMerror -contrast ASvfix_6 -a 2 -c 0 -setwdelay -sumdelays stxgrinder-sess -a EMerror -c ASvfix_6 -sf Subject_Lists/bw-group-error-5
# resample onto average7 # ( use version from doug's home directory with bugs corrected... )
/space/greve/1/users/greve/freesurfer/bin/mris_preproc --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --target average7 --hemi lh --sf Subject_Lists/bw-group-error-5 --a EMerror --c ASvfix_6 --out lh.ASvfix_8s_BwGroupError5.fsfast.mgh | tee -a logs/mris_preproc_BwGroupError5_ASvfix_8s_lh.log /space/greve/1/users/greve/freesurfer/bin/mris_preproc --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --target average7 --hemi rh --sf Subject_Lists/bw-group-error-5 --a EMerror --c ASvfix_6 --out rh.ASvfix_8s_BwGroupError5.fsfast.mgh | tee -a logs/mris_preproc_BwGroupError5_ASvfix_8s_rh.log
mkdir lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/ mkdir rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/
# lh mri_glmfit --really-use-average7 --y lh.ASvfix_8s_BwGroupError5.fsfast.mgh --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --glmdir lh.ASvfix_8s_BwGroupError5.fsfast.glmdir --surf average7 lh --C fsgd_and_contrast_files/BwGroupError5.mat --sim mc-full 5000 2 BwGroupError5_ASvfix_8s_lhMCSIM1 --fwhm 13.488420 | tee -a logs/glmfit_lh_MCSIM1.log mri_glmfit --really-use-average7 --y lh.ASvfix_8s_BwGroupError5.fsfast.mgh --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --glmdir lh.ASvfix_8s_BwGroupError5.fsfast.glmdir --surf average7 lh --C fsgd_and_contrast_files/BwGroupError5.mat --sim mc-full 5000 2 BwGroupError5_ASvfix_8s_lhMCSIM2 --fwhm 13.488420 | tee -a logs/glmfit_lh_MCSIM2.log mri_surfcluster --src lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.mgh --csd BwGroupError5_ASvfix_8s_lhMCSIM1-BwGroupError5.csd --csd BwGroupError5_ASvfix_8s_lhMCSIM2-BwGroupError5.csd --sum lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.clusterMCSIM.sum --ocp lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.clusterMCSIM.mgh --thmax infinity | tee -a logs/surfcluster_lh_BwGroupError5_AsvFix_mcsim.log # rh mri_glmfit --really-use-average7 --y rh.ASvfix_8s_BwGroupError5.fsfast.mgh --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --glmdir rh.ASvfix_8s_BwGroupError5.fsfast.glmdir --surf average7 rh --C fsgd_and_contrast_files/BwGroupError5.mat --sim mc-full 5000 2 BwGroupError5_ASvfix_8s_rhMCSIM1 --fwhm 13.488420 | tee -a logs/glmfit_rh_MCSIM1.log mri_glmfit --really-use-average7 --y rh.ASvfix_8s_BwGroupError5.fsfast.mgh --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --glmdir rh.ASvfix_8s_BwGroupError5.fsfast.glmdir --surf average7 rh --C fsgd_and_contrast_files/BwGroupError5.mat --sim mc-full 5000 2 BwGroupError5_ASvfix_8s_rhMCSIM2 --fwhm 13.488420 | tee -a logs/glmfit_rh_MCSIM2.log mri_surfcluster --src rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.mgh --csd BwGroupError5_ASvfix_8s_rhMCSIM1-BwGroupError5.csd --csd BwGroupError5_ASvfix_8s_rhMCSIM2-BwGroupError5.csd --sum rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.clusterMCSIM.sum --ocp rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.clusterMCSIM.mgh --thmax infinity | tee -a logs/surfcluster_rh_BwGroupError5_AsvFix_mcsim.log
# getting the correct map to display
# ( not yet done ) # # here is Frida's example # # mri_surfcluster --src lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.mgh --o lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_o.mgh --hemi lh --srcsubj average7 --thmin 1.3 # mri_surfcluster --src lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.mgh --csd BwGroup3b_ASevASc_6s_lh_MCSIM1-BwGroupError3b_NvSZ_contrast.csd --csd BwGroup3b_ASevASc_6s_lh_MCSIM2-BwGroupError3b_NvSZ_contrast.csd --thmax infinity --cwsig lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_cwsig.mgh
# mri_mask -T 1.3 lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_o.mgh lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_cwsig.mgh lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_ocwsig.mgh
--
Robert P. Levy, B.A. Research Assistant, Manoach Lab Massachusetts General Hospital Charlestown Navy Yard 149 13th St., Room 2656 Charlestown, MA 02129 email: levy@nmr.mgh.harvard.edu phone: 617-726-1908 fax: 617-726-4078
Hi Rob,
Yes you can run as many iterations as you want to in parallel. The monte carlo does take a long time to run, how many subjects are in your study?
Jenni
On Thu, 5 Jul 2007, Robert Levy wrote:
Hello,
This is probably a quick question. It seems to take a long time to generate monte carlo simulation data in mri_glmfit. Is it a bad idea to break the iterations into however many available seychelles nodes I can use? Is that equivalent to running the total number of iterations serially? Thanks, Rob
fpolli@nmr.mgh.harvard.edu wrote:
hi rob, see below for the answers to some of your questions.
Hi Dara and Frida,
The analysis is running. There is one part of Frida's example that I will need to emulate once this is done, but if what I have is correct, then the part that takes the most time will be done when I come in to work on Thursday 7/5.
Frida, sometime when you have a moment free, could you look over the commands I'm running and tell me if there is anything glaringly (or subtly) wrong? I have tried my best to run it in a way that is consistent with how you did it, but I think there could be things wrong, because I dont 100% understand everything about what these commands are doing, but getting there. There's nothing in these commands that looks like nonsense to me, but for example, I don't know if the smoothing value of 13.488420 is the same in our case.
to find this parameter, you want to open a file that's called mri_glmfit_log or something like that. it is created in the third step. however, it usually gives you a value b/w 12.5 and 14.5 and using one vs.the other does not create a huge diffrence. however, try to find the file i'm talking about, open it up, and you'll see.
I'm using a threshold of .01, and I
don't know if that would cause any parameters other than the threshold to need to change.
you should use whatever parameter you want your data displayed at. no that doesn't change anything else.
I also don't know why simulation data is generated
twice and then both input into the contrast,
only b/c it takes a long time to generate 10K simulations, so we broke it up into 2 x 5K. you can also use 1 x 10 k. but I'm following that
aspect of your example verbatim at least for now, in the commands that are now running. Also a default glmfit setting is to use dods (diff order diff slope).
dods is DEF. the one you want to use. it stands for different offset different slope - you don't want to assume that both groups have the same slope to their data.
Is that what we should be using? There is also a
doss (same slope) option. Ally, she is one of the RAs who I talked with yesterday about this, mentioned that dods is more unpredictable than doss.
Thanks for your help, Rob
good luck, f
#!/usr/local/bin/bash
################## # levy 7/3/07 # ##################
# Katy ran intergroupavg-sess on 9/20/05 # Last week we used FDR to correct for multiple comparisons on the results of the above contrast between groups, # but the resulting threshold was too conservative. # So now I am running the analysis with a cluster-based threshold using a Monte Carlo simulation # (simulate a null hypothesis with randomnes, and use it to generate a meaningful threshold to judge the results with) # In terms of the freesurfer implementation, this is a different process and does not build on the results of previous commands.
# so we start from the beginning # note: rerunning these 2 lines probably unnecessary, but mris_preproc was having problems I thought were due to faulty stxgrinder output # This problem led Doug to discover bugs in mris_preproc and also mkcontrast-sess, which he then fixed # As a result, I re-ran the following 2 commands, which didn't take very long anyhow.
mkcontrast-sess -analysis EMerror -contrast ASvfix_6 -a 2 -c 0 -setwdelay -sumdelays stxgrinder-sess -a EMerror -c ASvfix_6 -sf Subject_Lists/bw-group-error-5
# resample onto average7 # ( use version from doug's home directory with bugs corrected... )
/space/greve/1/users/greve/freesurfer/bin/mris_preproc --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --target average7 --hemi lh --sf Subject_Lists/bw-group-error-5 --a EMerror --c ASvfix_6 --out lh.ASvfix_8s_BwGroupError5.fsfast.mgh | tee -a logs/mris_preproc_BwGroupError5_ASvfix_8s_lh.log /space/greve/1/users/greve/freesurfer/bin/mris_preproc --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --target average7 --hemi rh --sf Subject_Lists/bw-group-error-5 --a EMerror --c ASvfix_6 --out rh.ASvfix_8s_BwGroupError5.fsfast.mgh | tee -a logs/mris_preproc_BwGroupError5_ASvfix_8s_rh.log
mkdir lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/ mkdir rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/
# lh mri_glmfit --really-use-average7 --y lh.ASvfix_8s_BwGroupError5.fsfast.mgh --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --glmdir lh.ASvfix_8s_BwGroupError5.fsfast.glmdir --surf average7 lh --C fsgd_and_contrast_files/BwGroupError5.mat --sim mc-full 5000 2 BwGroupError5_ASvfix_8s_lhMCSIM1 --fwhm 13.488420 | tee -a logs/glmfit_lh_MCSIM1.log mri_glmfit --really-use-average7 --y lh.ASvfix_8s_BwGroupError5.fsfast.mgh --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --glmdir lh.ASvfix_8s_BwGroupError5.fsfast.glmdir --surf average7 lh --C fsgd_and_contrast_files/BwGroupError5.mat --sim mc-full 5000 2 BwGroupError5_ASvfix_8s_lhMCSIM2 --fwhm 13.488420 | tee -a logs/glmfit_lh_MCSIM2.log mri_surfcluster --src lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.mgh --csd BwGroupError5_ASvfix_8s_lhMCSIM1-BwGroupError5.csd --csd BwGroupError5_ASvfix_8s_lhMCSIM2-BwGroupError5.csd --sum lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.clusterMCSIM.sum --ocp lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.clusterMCSIM.mgh --thmax infinity | tee -a logs/surfcluster_lh_BwGroupError5_AsvFix_mcsim.log # rh mri_glmfit --really-use-average7 --y rh.ASvfix_8s_BwGroupError5.fsfast.mgh --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --glmdir rh.ASvfix_8s_BwGroupError5.fsfast.glmdir --surf average7 rh --C fsgd_and_contrast_files/BwGroupError5.mat --sim mc-full 5000 2 BwGroupError5_ASvfix_8s_rhMCSIM1 --fwhm 13.488420 | tee -a logs/glmfit_rh_MCSIM1.log mri_glmfit --really-use-average7 --y rh.ASvfix_8s_BwGroupError5.fsfast.mgh --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --glmdir rh.ASvfix_8s_BwGroupError5.fsfast.glmdir --surf average7 rh --C fsgd_and_contrast_files/BwGroupError5.mat --sim mc-full 5000 2 BwGroupError5_ASvfix_8s_rhMCSIM2 --fwhm 13.488420 | tee -a logs/glmfit_rh_MCSIM2.log mri_surfcluster --src rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.mgh --csd BwGroupError5_ASvfix_8s_rhMCSIM1-BwGroupError5.csd --csd BwGroupError5_ASvfix_8s_rhMCSIM2-BwGroupError5.csd --sum rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.clusterMCSIM.sum --ocp rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.clusterMCSIM.mgh --thmax infinity | tee -a logs/surfcluster_rh_BwGroupError5_AsvFix_mcsim.log
# getting the correct map to display
# ( not yet done ) # # here is Frida's example # # mri_surfcluster --src lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.mgh --o lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_o.mgh --hemi lh --srcsubj average7 --thmin 1.3 # mri_surfcluster --src lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.mgh --csd BwGroup3b_ASevASc_6s_lh_MCSIM1-BwGroupError3b_NvSZ_contrast.csd --csd BwGroup3b_ASevASc_6s_lh_MCSIM2-BwGroupError3b_NvSZ_contrast.csd --thmax infinity --cwsig lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_cwsig.mgh
# mri_mask -T 1.3 lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_o.mgh lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_cwsig.mgh lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_ocwsig.mgh
--
Robert P. Levy, B.A. Research Assistant, Manoach Lab Massachusetts General Hospital Charlestown Navy Yard 149 13th St., Room 2656 Charlestown, MA 02129 email: levy@nmr.mgh.harvard.edu phone: 617-726-1908 fax: 617-726-4078
btw, which monte carlo are you running? The z or the t? The z is more conservative but it is *much* faster.
On Thu, 5 Jul 2007, Jenni Pacheco wrote:
Hi Rob,
Yes you can run as many iterations as you want to in parallel. The monte carlo does take a long time to run, how many subjects are in your study?
Jenni
On Thu, 5 Jul 2007, Robert Levy wrote:
Hello,
This is probably a quick question. It seems to take a long time to generate monte carlo simulation data in mri_glmfit. Is it a bad idea to break the iterations into however many available seychelles nodes I can use? Is that equivalent to running the total number of iterations serially? Thanks, Rob
fpolli@nmr.mgh.harvard.edu wrote:
hi rob, see below for the answers to some of your questions.
Hi Dara and Frida,
The analysis is running. There is one part of Frida's example that I will need to emulate once this is done, but if what I have is correct, then the part that takes the most time will be done when I come in to work on Thursday 7/5.
Frida, sometime when you have a moment free, could you look over the commands I'm running and tell me if there is anything glaringly (or subtly) wrong? I have tried my best to run it in a way that is consistent with how you did it, but I think there could be things wrong, because I dont 100% understand everything about what these commands are doing, but getting there. There's nothing in these commands that looks like nonsense to me, but for example, I don't know if the smoothing value of 13.488420 is the same in our case.
to find this parameter, you want to open a file that's called mri_glmfit_log or something like that. it is created in the third step. however, it usually gives you a value b/w 12.5 and 14.5 and using one vs.the other does not create a huge diffrence. however, try to find the file i'm talking about, open it up, and you'll see.
I'm using a threshold of .01, and I
don't know if that would cause any parameters other than the threshold to need to change.
you should use whatever parameter you want your data displayed at. no that doesn't change anything else.
I also don't know why simulation data is generated
twice and then both input into the contrast,
only b/c it takes a long time to generate 10K simulations, so we broke it up into 2 x 5K. you can also use 1 x 10 k. but I'm following that
aspect of your example verbatim at least for now, in the commands that are now running. Also a default glmfit setting is to use dods (diff order diff slope).
dods is DEF. the one you want to use. it stands for different offset different slope - you don't want to assume that both groups have the same slope to their data.
Is that what we should be using? There is also a
doss (same slope) option. Ally, she is one of the RAs who I talked with yesterday about this, mentioned that dods is more unpredictable than doss.
Thanks for your help, Rob
good luck, f
#!/usr/local/bin/bash
################## # levy 7/3/07 # ##################
# Katy ran intergroupavg-sess on 9/20/05 # Last week we used FDR to correct for multiple comparisons on the results of the above contrast between groups, # but the resulting threshold was too conservative. # So now I am running the analysis with a cluster-based threshold using a Monte Carlo simulation # (simulate a null hypothesis with randomnes, and use it to generate a meaningful threshold to judge the results with) # In terms of the freesurfer implementation, this is a different process and does not build on the results of previous commands.
# so we start from the beginning # note: rerunning these 2 lines probably unnecessary, but mris_preproc was having problems I thought were due to faulty stxgrinder output # This problem led Doug to discover bugs in mris_preproc and also mkcontrast-sess, which he then fixed # As a result, I re-ran the following 2 commands, which didn't take very long anyhow.
mkcontrast-sess -analysis EMerror -contrast ASvfix_6 -a 2 -c 0 -setwdelay -sumdelays stxgrinder-sess -a EMerror -c ASvfix_6 -sf Subject_Lists/bw-group-error-5
# resample onto average7 # ( use version from doug's home directory with bugs corrected... )
/space/greve/1/users/greve/freesurfer/bin/mris_preproc --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --target average7 --hemi lh --sf Subject_Lists/bw-group-error-5 --a EMerror --c ASvfix_6 --out lh.ASvfix_8s_BwGroupError5.fsfast.mgh | tee -a logs/mris_preproc_BwGroupError5_ASvfix_8s_lh.log /space/greve/1/users/greve/freesurfer/bin/mris_preproc --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --target average7 --hemi rh --sf Subject_Lists/bw-group-error-5 --a EMerror --c ASvfix_6 --out rh.ASvfix_8s_BwGroupError5.fsfast.mgh | tee -a logs/mris_preproc_BwGroupError5_ASvfix_8s_rh.log
mkdir lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/ mkdir rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/
# lh mri_glmfit --really-use-average7 --y lh.ASvfix_8s_BwGroupError5.fsfast.mgh --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --glmdir lh.ASvfix_8s_BwGroupError5.fsfast.glmdir --surf average7 lh --C fsgd_and_contrast_files/BwGroupError5.mat --sim mc-full 5000 2 BwGroupError5_ASvfix_8s_lhMCSIM1 --fwhm 13.488420 | tee -a logs/glmfit_lh_MCSIM1.log mri_glmfit --really-use-average7 --y lh.ASvfix_8s_BwGroupError5.fsfast.mgh --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --glmdir lh.ASvfix_8s_BwGroupError5.fsfast.glmdir --surf average7 lh --C fsgd_and_contrast_files/BwGroupError5.mat --sim mc-full 5000 2 BwGroupError5_ASvfix_8s_lhMCSIM2 --fwhm 13.488420 | tee -a logs/glmfit_lh_MCSIM2.log mri_surfcluster --src lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.mgh --csd BwGroupError5_ASvfix_8s_lhMCSIM1-BwGroupError5.csd --csd BwGroupError5_ASvfix_8s_lhMCSIM2-BwGroupError5.csd --sum lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.clusterMCSIM.sum --ocp lh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.clusterMCSIM.mgh --thmax infinity | tee -a logs/surfcluster_lh_BwGroupError5_AsvFix_mcsim.log # rh mri_glmfit --really-use-average7 --y rh.ASvfix_8s_BwGroupError5.fsfast.mgh --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --glmdir rh.ASvfix_8s_BwGroupError5.fsfast.glmdir --surf average7 rh --C fsgd_and_contrast_files/BwGroupError5.mat --sim mc-full 5000 2 BwGroupError5_ASvfix_8s_rhMCSIM1 --fwhm 13.488420 | tee -a logs/glmfit_rh_MCSIM1.log mri_glmfit --really-use-average7 --y rh.ASvfix_8s_BwGroupError5.fsfast.mgh --fsgd fsgd_and_contrast_files/BwGroupError5.fsgd --glmdir rh.ASvfix_8s_BwGroupError5.fsfast.glmdir --surf average7 rh --C fsgd_and_contrast_files/BwGroupError5.mat --sim mc-full 5000 2 BwGroupError5_ASvfix_8s_rhMCSIM2 --fwhm 13.488420 | tee -a logs/glmfit_rh_MCSIM2.log mri_surfcluster --src rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.mgh --csd BwGroupError5_ASvfix_8s_rhMCSIM1-BwGroupError5.csd --csd BwGroupError5_ASvfix_8s_rhMCSIM2-BwGroupError5.csd --sum rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.clusterMCSIM.sum --ocp rh.ASvfix_8s_BwGroupError5.fsfast.glmdir/BwGroupError5/sig.clusterMCSIM.mgh --thmax infinity | tee -a logs/surfcluster_rh_BwGroupError5_AsvFix_mcsim.log
# getting the correct map to display
# ( not yet done ) # # here is Frida's example # # mri_surfcluster --src lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.mgh --o lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_o.mgh --hemi lh --srcsubj average7 --thmin 1.3 # mri_surfcluster --src lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.mgh --csd BwGroup3b_ASevASc_6s_lh_MCSIM1-BwGroupError3b_NvSZ_contrast.csd --csd BwGroup3b_ASevASc_6s_lh_MCSIM2-BwGroupError3b_NvSZ_contrast.csd --thmax infinity --cwsig lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_cwsig.mgh
# mri_mask -T 1.3 lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_o.mgh lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_cwsig.mgh lh.ASevASc_6s_BwGroup3b.fsfast.glmdir/BwGroupError3b_NvSZ_contrast/sig.clusterMCSIM_ocwsig.mgh
--
Robert P. Levy, B.A. Research Assistant, Manoach Lab Massachusetts General Hospital Charlestown Navy Yard 149 13th St., Room 2656 Charlestown, MA 02129 email: levy@nmr.mgh.harvard.edu phone: 617-726-1908 fax: 617-726-4078
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
freesurfer@nmr.mgh.harvard.edu
-
Doug Greve -
Jenni Pacheco -
Robert Levy