Dear PLS Experts,
Could you comment/suggest on the following setup of Structural PLS analysis to assess the grey matter differences between two groups. I have two groups (controls and patients) with grey matter images (segmented, modulated and smoothed using VBM(SPM) preprocessing). As described in the PLS manual, I have set up grey matter images as the only condition and create the "group_STRUCTURAL.mat" file for both groups. Then I run a "Mean-Centering PLS" with the two groups and got two LV (the first one reflect the difference of GM with p>.98). so, I got a strange statistical map "pixelized" with a bootstrap scale ranged between +/- 4000. Where do you think is the mistake? In the setup of the structural PLS analysis? Or probably the grey matter images have to be processed in a specific way? I hope that I have been clear,
Many thanks,
Mbahri
Take a look at:
http://www.rotman-baycrest.on.ca/pls/faq.txt
It tells us:
"In order to run Mean-Centering PLS, you need to have at least 2 conditions. However, if you have only 1 condition and you have more than 1 groups, our PLS will reconstruct datamat so that it will become single group with multiple conditions."
That is why you got confusing statistical map.
In one word, you need to have a 2nd condition, or run other PLS analysis methods instead of "Mean-Centering" PLS.
Hi Jimmy,
Thanks a lot for your helpful answer. Do you think that I can use the "Non-Rotated Task PLS" method with 1 -1 as contrast to assess the grey matter differences? of course with same setup (2 groups and one condition "GM").
Thanks,
mbahri
It looks okay for me. However, I cannot provide you further answer if it does not work.
When you run Non-Rotated Task PLS, you can only provide 1 value for each group, since you only have 1 condition. You can input 2 contrast like this:
Group1 Contrast 1: 1
Group1 Contrast 2: -1
Group2 Contrast 1: -1
Group2 Contrast 2: 1
Try it, and let us know if it works.
I checked with Randy. In your case (with two groups and one condition), you should be able to use Mean-Centering PLS to do your analysis, although I cannot provide you with any satisfactory answer. I understand that you got a strange statistical map, which could be caused by your raw data, or something else that beyond my knowledge, and I am really sorry that I told you not to use Mean-Centering PLS.
I forgot to mention that you can select a different Mean-Centering Type option than the default - i.e. 1 instead of 0.
For details about Mean-Centering Type feature, please take a look at User Guide for PLS Applications, see web link below:
http://research.baycrest.org/pls/userguide.htm#meancentering_type
If you can not see Mean-Centering Type pull-down menu in PLS analysis window, it means that the version of PLS that you are using needs to be updated. Please download the latest version of PLS applications from the web link below:
http://research.baycrest.org/pls/source
Hi Jimmy,
I run all possible Mean-Centering PLS types (0, 1, 2, 3) and I got the same strange statistical map as results. When Mean-Centering PLS 1, 2, or 3 is used the following information appear on the matlab window : “Because you are running single condition Mean-Centering PLS, input Mean-Centering Type has to reset to 0.”. I do not think that there is a problem with my data.
However, Non-rotated PLS Task method both with one contrast (g1: 1, g2: -1) or two contrasts (g1: 1, -1; g2: -1, 1) provides the same reasonable results (normal statical maps with bootstrap scale between -6 and 6).
When I run the Non-rotated PLS method with the two contrasts the following warnings appear: “Warning 1: Contrasts are not linear independent” and “Warning 2: Contrasts are not orthogonal”.
It would be very helpful if you could have a look at the results. I can send you some screen captures showing the results, whether allowed by the policy of PLS forum.
Many thanks
mbahri
Hi..
Pixelized Bootstrap --> how many bootstrap samples did you run? I can replicate a pixelized bootstrap image using a silly n=3/group with only 7-10 bootstrap samples.
You can quickly check to see if the analysis "looks right" by showing the BrainLV, with no thresholding -
If you used no thresholding for your modulated segmentation results, you should see something like this. If you have thresholded, the white matter should be removed.
OK.. so the jpg wasn't allowed... here is the link to the jpg on dropbox:
https://www.dropbox.com/s/zol4aoolgvjwdr2/StructModule_BrainLV.jpg
Hi,
Number of subjects per group is : 19 for group 1 and 31 for group 2. All analysis were done with 100 iterations for bootstrap and 600 permutation. Screen captures showing results are on the dropbox ( the following link)
https://www.dropbox.com/sh/t7fma4n15pce6vq/mM9ax0Dyiw
Mohamad..
For the Mean-Centred analysis: what are the singular values?
can you also put up a non-thresholded bootstrap image from the mean-centred and non-rotated analyses - I'd like to see if you have a reasonable pattern of bootstrap ratios -
nancy
Hi Nancy,
Plots of singular values for Mean_Centred analysis as well as the non-thresholded bootstrap image from mean-centred and non-rotated analysis are in the following dropbox link.
https://www.dropbox.com/sh/t7fma4n15pce6vq/mM9ax0Dyiw
The results obtained by non-rotated analysis corroborate well with the results obtained using an univariate analysis (SPM).
Is it possible to control for nuisance covariables (e.g. Age, sex ...) when using Non-Rotated behvioral PLS?
Thanks a lot for your help,
Mohamed
Mohamad..
Your LV1 salience looks OK for the mean-centred analysis.. the range for the bootstrap ratios do not - we're checking that out.
I've made a note on your bootstrap figure - can you redo that figure for me?
https://www.dropbox.com/s/urs432l4ngn5g2i/MeanCenterPLS_njl.png
nancy
We just made a change to the computation of "Single Condition Task PLS". Therefore, please download the latest version and run your analysis again from:
http://research.baycrest.org/pls/source
You will expect the following information: "Because you are running single condition Task PLS, input Mean-Centering Type has to reset to 1."
Hi Nancy and Jimmy,
The modified version of Mean-Centering PLS works now. The obtained bootstrap image is similar to the one obtained with the Non rotated PLS task (see figures in the dropbox link).
https://www.dropbox.com/sh/t7fma4n15pce6vq/mM9ax0Dyiw
I also added the bootstrap figure (obtained with Mean-Centering before modification) modified according to Nancy's note.
Is it possible to control for the nuisance co-variables (e.g. Age, Sex …) when using the Non Rotated behavioral PLS?
Many thanks,
Mohamed,
Hi Mohamad..
if the speckled image is from the updated gui, there are still issues, if not, you look good to go..
WRT nuisance variables: we do not provide that capability in the gui - you should use your favourite program to regress those variables against the data and input the regressed data into the analysis..
Nancy
The speckled image is not from the updated gui.
WRT nuisance variables: I do not have any program for that. Could you recommend one?
Baycrest is an academic health sciences centre fully affiliated with the University of Toronto
Privacy Statement - Disclaimer - © 1989-2024 BAYCREST HEALTH SCIENCE. ALL RIGHTS RESERVED