Hi Jessica

Many thanks for your answers. 
I would like to discuss again the second question related to the MKL model. I think I wasn't clear enough here. 

I have task-related fMRI data in the form of processed beta-images including brain responses related to fear. 

Now say I have two different questionnaires measuring fear. As a form of "neural validation" I would like to know which questionnaire is more associated with the fear-related brain activity (which questionnaire and scores can be better predicted from brain activity data). 

For this purpose I would like to use an MKL machine using the two questionnaires as two modalities (same beta images but different regression targets from the two questionnaires) and use the option "use one kernel per modality". 
Furthermore, I use a feature set based on voxels related to fear (Neurosynth mask based on the term fear). 

By doing this I thought I could answer the question which questionnaire score is better predictable by the respective brain activity data by looking at modality weight. 

If this approach is wrong do you see another option ? 

Many thanks for your help! 

best, mike