Partial least squares as a target-directed structure-seeking technique

Partial least square (PLS) has enjoyed considerable success in a variety of applications. One of the reasons it is so popular is that it can be applied in many different ways, with a host of different endpoints of interest for the analysis. In chemometrics, the focus has traditionally been on calibration or regression. There has been very little discussion about how to use PLS as a target-directed structure-seeking technique in kinetic studies. It turns out that this is a very natural use of PLS, one that emerges as preferable to the standard approach that employs a least-squares fit of selected component directions to the kinetic target. In this paper, we explore this comparison within the context of brain mapping studies employing functional magnetic resonance imaging (fMRI).