Chapter 17 Partial Least Squares (Pls) for the Prediction of Real-Life Performance from Laboratory Results

Abstract

This chapter predicts the real-life result of new test objects from the corresponding laboratory data only. The model involves two or more independent blocks of data that are to be correlated with one block of dependent data. It is assumed that each block describes the same number of objects that are represented in identical order by the rows of the blocks. The variables are represented by the columns of the various blocks in the model. They usually differ in nature and in number from one block to another. The method is illustrated by means of an application from the development of novel pharmaceutical compounds in the field of psychiatry. The solution of partial least squares (PLS) is described. This approach involves the calculation of latent variables, one for each independent block. Prior to performing the PLS analysis, a number of preprocessing steps to the various blocks of data are applied.