O2PLS uses linear regression to divide the systematic variation in X and Y into three parts; one part with joint X–Y covariation, i.e. related to both X and Y, one part of X with Y-orthogonal variation and one part of Y with X-orthogonal variation.
All of the investigated pre-treatment methods removed an additive baseline as expected. In the analysis of raw and differentiated data variation associated with the baseline was found in the Y-orthogonal part of X. Orthogonal information was also found in Y, which suggests that this pre-processing procedure not only removed variation. This would have been more difficult to detect without the O2PLS model since both raw and differentiated data must be analysed simultaneously.
Development of a knowledge based strategy with OPLS methodology is an important step towards eliminating trial and error approaches to pre-processing.