Abstract
This article introduces a generalization of the partial least squares regression (PLS). Transforming the predictors by means of spline functions is a useful way to extend PLS into nonlinearity and to obtain a multiresponse additive model. We describe both statistical and computational aspects of this new method, termed additive splines partial least squares (ASPLS). The performance of ASPLS compared with other PLS methods is illustrated with chemical and physiological applications.