Partial Least Squares Regression, also called Projection to Latent Structures, can be applied to establish a predictive model, even if the features are highly correlated. This makes PLS an attractive method for QSAR. The goal of PLS is to establish a relationship between the two matrices X and Y. The procedure is as follows: first, the principal components for X and Y are calculated separately (cf. PCA). The scores of the matrix X are then used for a regression model to predict the scores of Y which can then be used to predict Y. A commonly used algorithms for calculating PLS is SIMPLS.
Ulrike Burkard, Dec. 12., 2002