Back to PLS HelpPosted on 10/24/14 13:36:42
Number of posts: 36
Greetings PLS experts,
I had a question in regards to conducting seed PLS based on the results of Task PLS. My understanding (which very well may be wrong...) from reading different threads on this forum and papers , it is common to do a task PLS with the conditions of the experiment, then based on areas that were found to be significalty differently activated in one condition vs the other, conduct a seed analysis based on some of these areas.
However, I'm confused as to the analysis itself. For example, if you have 3 regions of interest based on the Task PLS results, are Seed PLS seed analyses done for each regions (as in one Seed PLS analysis ran for one region of interest), or is there some method where a single analysis is done with all regions of interest included.
For example, in the paper "Age-related dedifferentiation and compensatory changes in the functional network underlying face processing" by Burianova, Lee, Grady and Moscovitch, they seemed to have ran a single seed analysis (from my intepretation) with multiple regions, producing just a single LV that reflected connectivity for each of the regions used. However, I believe there are papers which also conduct one seed PLS for each region, although I may be incorrect.
I am just confused as to when having multiple regions of interest, doing a single seed analysis for one region vs doing a single analysis with all regions, and if one is more appropriate than the other. Any help with this would be greatly appreciated.
Thank you!
Replies:
Posted on 10/24/14 14:51:09
Number of posts: 394
Hi Paul - there is no strict rule on when to do seed PLS with one region vs multiple. Part of the choice will depend on the balance of being comprehensive versus interpretation. Multiple regions in a single analysis allows you to determine where their pattern of functional connectivity is similar and different. Doing separate analyses does not. However, a multiple seed analysis can get tricky when there are a number of groups/conditions as the dimensionality can get overwhelming. My suggestion would be to tailor the approach based on the question. If that alone does not help, you can do both, starting with multiple regions first to get an idea of the result space and follow up with single region analysis as a sort of post-hoc.
I hope that helps
Randy
Posted on 10/24/14 15:08:13
Number of posts: 36
Hi Paul - there is no strict rule on when to do seed PLS with one region vs multiple. Part of the choice will depend on the balance of being comprehensive versus interpretation. Multiple regions in a single analysis allows you to determine where their pattern of functional connectivity is similar and different. Doing separate analyses does not. However, a multiple seed analysis can get tricky when there are a number of groups/conditions as the dimensionality can get overwhelming. My suggestion would be to tailor the approach based on the question. If that alone does not help, you can do both, starting with multiple regions first to get an idea of the result space and follow up with single region analysis as a sort of post-hoc.
I hope that helps
Randy
Hi Randy,
thanks for the helpful response. Sorry for following with another question, but in regards to conducting a single seed analysis for multiple regions, can it be done through the PLS GUI? Would you essentially do the multiple voxel extraction for each of the areas of interest, combine them into a single text file, and then use that as the behavioural text file?
Thanks again.
Posted on 10/24/14 15:28:34
Number of posts: 394
You can indeed do it in the GUI, but as you indicated you need to combine all the seed voxels into a single file (1 column per seed) and read it in. This also makes it easier for the follow-up analyses if you want to select a subset of voxels, as you can read in the same text file and then select only those seeds you need.
Posted on 10/24/14 16:46:55
Number of posts: 36
You can indeed do it in the GUI, but as you indicated you need to combine all the seed voxels into a single file (1 column per seed) and read it in. This also makes it easier for the follow-up analyses if you want to select a subset of voxels, as you can read in the same text file and then select only those seeds you need.
Thank you again Randy.
Just as a follow up to creating the single text file with multiple seed regions (1 seed with its own column), is there a way to go about extracting the voxel data in the previous step from only a single condition, or would this just be done from the resulting .mat file (curious since the paper I mentioned before, where they looked at multiple seed regions in a single Seed PLS analysis, only focused on a single task with the resulting figures displaying this single task network associated with the selected seeds ). Thanks again for all the help.
Posted on 10/24/14 19:07:34
Number of posts: 394
Thank you again Randy.
Just as a follow up to creating the single text file with multiple seed regions (1 seed with its own column), is there a way to go about extracting the voxel data in the previous step from only a single condition, or would this just be done from the resulting .mat file (curious since the paper I mentioned before, where they looked at multiple seed regions in a single Seed PLS analysis, only focused on a single task with the resulting figures displaying this single task network associated with the selected seeds ). Thanks again for all the help.
Its probably best to use the entire dataset for the seed extraction and then you can edit the text file afterwards to create the condition-specific ones you want. The extracted seeds come out as subjects within conditions within groups.
Login to reply to this topic.
