Back to PLS HelpBehavioural correlation values and Brainscores
Posted on 10/14/10 11:40:20
Number of posts: 7
Hi there!
I am using PLS to look at grey matter data, with 3 groups and 6 behavioural measures. Two questions have popped up!
Firstly, I've noticed that if I load up the behavlv output in matlab, the correlation values are different to the r values displayed in the 'brain scores plot for behaviour analysis'... can you possibly explain what the difference between these outputs are please?
Also, I am a little confused by my brain scores ranging from negative through to positive numbers for the same LV. Is this due to a normalisation process? I'm trying to make inference to how brain score relates to GM vol so would be great to have a better handle on this.
Thank you! All the best, Anna
I am using PLS to look at grey matter data, with 3 groups and 6 behavioural measures. Two questions have popped up!
Firstly, I've noticed that if I load up the behavlv output in matlab, the correlation values are different to the r values displayed in the 'brain scores plot for behaviour analysis'... can you possibly explain what the difference between these outputs are please?
Also, I am a little confused by my brain scores ranging from negative through to positive numbers for the same LV. Is this due to a normalisation process? I'm trying to make inference to how brain score relates to GM vol so would be great to have a better handle on this.
Thank you! All the best, Anna
Replies:
Untitled Post
I'm Online
Posted on 10/14/10 12:14:16
Number of posts: 291
I just tested a couple of small data, and the correlation values are the same as the r values. Is there any chance that you could upload all your your .mat file along with the behavior text file, so I can try to duplicate the problem? If so, please send me the link.
Untitled Post
Posted on 10/21/10 05:31:01
Number of posts: 7
Hi Jimmy,
Thanks for your quick response!
My colleague here in Cambridge has helped me with the correlation values mix up, thank you! I was getting two very distinct things mixed up!!!
I wonder if you could shine some light on the second question in my original message, thank you!
All the best,
Anna
Thanks for your quick response!
My colleague here in Cambridge has helped me with the correlation values mix up, thank you! I was getting two very distinct things mixed up!!!
I wonder if you could shine some light on the second question in my original message, thank you!
All the best,
Anna
Untitled Post
I'm Online
Posted on 10/21/10 08:41:49
Number of posts: 229
Hi Anna..
The brainscores are the dot-product of the LV saliences and the subject's actual data.
To take the example of a positive brainscore/behaviour correlation:
If the LV is mostly positive saliences, you will likely find that your brainscores are also positive for all subjects.
In the case of mixed positive and negative saliences, you may find both positive and negative brainscores. This means that for subjects with negative brainscores, their gray matter probabilities in regions of negative saliences are dominating.
To help you interpret the findings more accurately, I would suggest that you make some GM vs behaviour scatter plots. You should extract some voxels from regions of positive and negative saliences using the 'multiple voxels extraction' module in the GUI.
Also to note, (and this is more applicable to VBM type analyses than to functional analyses):
I urge you to look at the unthresholded brainLV (not the bootstrap ratios) to be sure that you do not have residual registration errors influencing your results. These will show up as nice swaths of, say, positive saliences, surrounded on one or both sides by negative saliences. If your positive and negative saliences are 'focal', then registration error is less likely to be influencing your results.
cheers,
Nancy
The brainscores are the dot-product of the LV saliences and the subject's actual data.
To take the example of a positive brainscore/behaviour correlation:
If the LV is mostly positive saliences, you will likely find that your brainscores are also positive for all subjects.
In the case of mixed positive and negative saliences, you may find both positive and negative brainscores. This means that for subjects with negative brainscores, their gray matter probabilities in regions of negative saliences are dominating.
To help you interpret the findings more accurately, I would suggest that you make some GM vs behaviour scatter plots. You should extract some voxels from regions of positive and negative saliences using the 'multiple voxels extraction' module in the GUI.
Also to note, (and this is more applicable to VBM type analyses than to functional analyses):
I urge you to look at the unthresholded brainLV (not the bootstrap ratios) to be sure that you do not have residual registration errors influencing your results. These will show up as nice swaths of, say, positive saliences, surrounded on one or both sides by negative saliences. If your positive and negative saliences are 'focal', then registration error is less likely to be influencing your results.
cheers,
Nancy
Untitled Post
Posted on 10/22/10 12:00:07
Number of posts: 7
thank you very much for this!
Login to reply to this topic.
