Multivariate QSAR

• Autor(a) principal: Ferreira,Márcia M. C.
• Data de Publicação: 2002
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Journal of the Brazilian Chemical Society (Online)
• Texto Completo: http://old.scielo.br/scielo.php?script=sci_arttext&pid=S0103-50532002000600004
• topic: chemometrics; principal component analysis; partial least squares; SIMCA; KNN

In this work, the chemometric techniques most frequently used in QSAR (quantitative structure-activity relationships) studies are reviewed. They are introduced in chronological order, beginning with Hansch analysis and the exploratory data analysis methods of principal components and hierarchical clustering (PCA and HCA). Principal component regression and partial least squares regression methods (PCR and PLS) are discussed, followed by the pattern recognition methods (KNN and SIMCA). Different applications are presented to illustrate these chemometric techniques. The methodology used for regression in 3D-QSAR is presented (unfolding PLS). Finally, the higher order method called Multilinear PLS, already used in analytical chemistry but not yet explored by the QSAR community, is introduced. This method maintains the multiway structure of the data and has several advantages over bilinear PLS including speed in calculation, simplicity and stability, since the number of parameters to be estimated can be greatly reduced.