Partial Least Squares (PLS) is a statistical technique which helps to analyse covariance structures of the predictors in relation to the outcome(s) of interest. In its raw form, PLS is a means of carrying out dimension reduction whilst facilitating the exchange of information between the predictors and the outcome(s). Because of this key ability, one can develop a model to predict the outcome(s) in terms of the most influential predictors determined by the PLS algorithm. Particularly in biomedical research, PLS is a useful tool for analysing high-dimensional biological datasets – namely OMICs data.
OMIC data measures the expression or abundance of molecules in biological systems. Therefore, studies use OMIC data in an effort to develop our current understanding of systems biology with a focus on molecular signatures of the organisms. Specifically, PLS proved useful for the challenging task of OMICs data integration where we can analyse data from across OMIC platforms (e.g. transcriptome, proteome etc) and reveal key molecular components of complex mechanisms.
The core abilities of PLS rely on finding the solution to an objective function or optimisation problem. Many versions of PLS appear in the literature because there are numerous variants of the optimisation problem, the techniques used to solve them; and the outputs of the algorithm. As a result, there is some degree of inconsistency in the terminology used in the literature which may confuse people reading from multiple sources.