Back to PLS HelpPosted on 10/25/15 00:18:40
Number of posts: 12
Hi everyone,
I do apologize in advance for asking what will be a very basic question, but after reading the methods/result interpretation of various PLS papers, I still feel as though I'm hitting my head against a brick wall. I'm primarily basing these questions based on PLS results from the Campbell et al., (2012 - Ages differences in the frontoparietal...) and Grady et al., (2010 - A multivariate analysis of age-related differences...) papers, which have multiple tasks and multiple groups (young vs old), which I think is causing some confusion for myself when interpretating the brain score bar graphs with the LVs.
My confusion is primarily based on the interpretation of the (negative or postive) brain scores/weighted salience in relation to the related LV voxel map. For example, let's say there is a PLS analysis done with 2 conditions, one of which was associated with positive brain scores (condition A), and the other with negative brain scores (condition B) and on the LV map, both positive and negative BSR voxels are found.
From my understanding, it wouldn't be enough to simply say positive BSR regions are associated with condition A, and negative BSR regions with condition B. Instead, the interpretation could be that positive BSR regions (which relate to condition A due to positive brain scores), are regions which showed greater activity for condition A compared to B, and the opposite would be true for negative BSR regions, that these regions were those that showed greater activity during condition B compared to condition A (I think this addresses the concept of clusters being positively or negatively correlated with the LV pattern?). Would this interpretation be correct?
A second question comes in the case where we have the same 2 conditions (A and B, which respectively are associated with positive and negative brain scores), but on the LV map, only have one of positive or negative BSR values. For example, if we had those 2 conditions, but only positive BSR voxels on the LV map, would it be correct to state that the positive BSR voxels were regions for which condition A was greater than the negative brain scored condition B (and that these same regions showed LESS activity for condition B compared to condition A - due to the negative brain score of condition B).
However, because there were no negative BSR voxels, there were no regions that showed greater activity for condition B (negative brain score) compared to condition A (positive brain score)? Would this be correct as well?
I'm sorry for the litany of questions littered throughout, but any response that could help clarify these questions would be absolutely wonderful. Thank you.
Replies:
Posted on 10/25/15 09:38:14
Number of posts: 394
Let me try to address your questions with some examples. Lets assume we are talking about functional imaging data where voxel values are expressed as %change from some reference point.
Scenario 1 where you have positive and negative saliences:
Voxel 1 Voxel 2 Voxel 3
Mean Condition A 10 5 0
Mean Condition B 5 10 -5
Salience (arbitrary) 0.5 -0.5 0.5
In this case there are positive and negative saliences. Voxel 1 is positive because the mean is higher in A vs B, voxel 2 is negative because its mean is higher is B vs A, and Voxel 3 is positive because its mean is higher in A vs B. Note that in Voxel 3, the difference between conditions comes because the voxel is less 'deactivated' in A
Now for the second example:
Scenario 2 where you have ony positive saliences:
Voxel 1 Voxel 2 Voxel 3
Mean Condition A 10 5 0
Mean Condition B 5 5 -5
Salience (arbitrary) 0.5 0 0.5
In this case Voxels 1 and 2 have positive saliences for the same reason as above.
The important point is how the scores from these are calculated. The scores are the multiplication of the raw data by the salience. If the raw data allow for negative values (deactivation) then the scores can be positive and negative. This would be the case for voxel 3 in both scenarios. The degree to which a person shows activation vs deactivation in a task will push the score to be more positive or more negative.
The final, and most important point is that you cannot interpret the saliences/BSR or scores without looking at the actual means that went into the analysis. Only after you look at the means can you interpret what the saliences and LV image reflects.
hope this is clear
Randy
Posted on 10/25/15 16:33:05
Number of posts: 12
Hi Randy,
thank you very much for the reply. When you refer to the means, are you referring to the mean brain scores that are commonly reported in papers along with the LV pattern, or the brain score of each individual subject (as found in the design score and LV plot graph in the plsgui...or I may be completely off). Thank you again.
Posted on 10/25/15 17:05:55
Number of posts: 394
the means in the examples are means for each voxel
Login to reply to this topic.
