Statistical Basis of Regression and Partial Least-Squares Analysis

This chapter will develop the statistical background necessary for quantitative structure-activity studies on modest-size data sets of related molecules: usually no more than a hundred analogs. Extension of this background to larger, more diverse data sets will be discussed in Chapter 11. The discussions that follow will develop linear regression analysis in some detail.[1,2] Once it is understood, it can be generalized to other multivariate statistical methods, in particular, partial least-squares (PLS) regression.