Back to PLS Helpmultiblock Task PLS
Posted on 02/02/07 12:16:32
Number of posts: 100
Author: Jennifer Labus (149.142.243.---)
Date: 11-29-06 18:03
A multiblock task pls is described in the manual. For clarification, does this analysis create a cross-block covariance matrix with the helmert design contrast and then correlate this with the seed by creating a cross-correlation matrix and then decompose this? I am not clear what the advantage of this multiblock is? Can you provide some further clarification?
Date: 11-29-06 18:03
A multiblock task pls is described in the manual. For clarification, does this analysis create a cross-block covariance matrix with the helmert design contrast and then correlate this with the seed by creating a cross-correlation matrix and then decompose this? I am not clear what the advantage of this multiblock is? Can you provide some further clarification?
Replies:
Re: multiblock Task PLS
Posted on 02/02/07 12:16:50
Number of posts: 100
Author: Randy (---.rotman-baycrest.on.ca)
Date: 11-30-06 17:33
The idea of the multiblock was presented in:
McIntosh, AR, Lobaugh, NJ, Cabeza, R, Bookstein, FL , & Houle, S., (1998). Convergence of neural systems processing stimulus associations and coordinating motor responses. Cerebral Cortex, 8, 648-659
but at that time the algorithm only identified one latent variable depicting the overlap between task effects and behaviour correlations.
Multiblock is a combination of task PLS and behaviour PLS and is meant to identify spatiotemporal patterns that simultaneously reflect activity differences and relate to some aspect of behaviour. It presently identifies any overlap that exists, plus the unique dimensions.
The algorithm works by computing the grand mean-centred data matrix averaged within condition (as you would for a task PLS) and the correlation matrix between your behaviour measure and activity within task. For clarity, lets assume you have v-voxels with t-time points and 4 tasks. The mean-centring and correlation process gives you two 4 by v*t matrices. The row of each matrix is normalized to unit-length to remove scaling differences. They are then stacked one on top of another to get an 8 by v*t matrix, which is subjected to SVD. The rest of the process is the same as the usual PLS, except that you get both design effects and brain-behaviour effects to evaluate.
While the procedure works well in practice, I do admit it is complicated and may be tough to interpret. We are working on ways to make it easier to relate to task and behaviour PLS done independently, which is on our ToDo list. Nevertheless, it does produce some fascinating results.
Date: 11-30-06 17:33
The idea of the multiblock was presented in:
McIntosh, AR, Lobaugh, NJ, Cabeza, R, Bookstein, FL , & Houle, S., (1998). Convergence of neural systems processing stimulus associations and coordinating motor responses. Cerebral Cortex, 8, 648-659
but at that time the algorithm only identified one latent variable depicting the overlap between task effects and behaviour correlations.
Multiblock is a combination of task PLS and behaviour PLS and is meant to identify spatiotemporal patterns that simultaneously reflect activity differences and relate to some aspect of behaviour. It presently identifies any overlap that exists, plus the unique dimensions.
The algorithm works by computing the grand mean-centred data matrix averaged within condition (as you would for a task PLS) and the correlation matrix between your behaviour measure and activity within task. For clarity, lets assume you have v-voxels with t-time points and 4 tasks. The mean-centring and correlation process gives you two 4 by v*t matrices. The row of each matrix is normalized to unit-length to remove scaling differences. They are then stacked one on top of another to get an 8 by v*t matrix, which is subjected to SVD. The rest of the process is the same as the usual PLS, except that you get both design effects and brain-behaviour effects to evaluate.
While the procedure works well in practice, I do admit it is complicated and may be tough to interpret. We are working on ways to make it easier to relate to task and behaviour PLS done independently, which is on our ToDo list. Nevertheless, it does produce some fascinating results.
Login to reply to this topic.
