The conditions and methods for constructing reliable QSARs are revised in relation to each component of a QSAR study: the selection of a training set out of a QSAR compatible series, the collection of data pertinent to the descriptors matrix (X) and to the effects matrix (Y), the analysis of data to connect X to Y by a regression model, and the validation of the model. In discussing these conditions, attention is given to the constraints that arise from the theoretical foundation of QSARs as analogy models of local validity and to the complexity and limited knowledge about the mechanisms of action. Hence, emphasis is placed on the need and importance to adopt multivariate methods for dealing with (1) the characterization of the structures, (2) the selection of a representative set of training compounds, and (3) analysis of the data. It is finally shown that the same integrated multivariate approach applies to the modeling of biological activities and other properties - chemical and biological - as well as to the modeling of correlations between batteries of data. The role of QSAR in risk assessment is addressed in the second part of the article. The framework of a strategy for an efficient screening assessment of toxic substances through the modeling of their exposure and toxicity-related properties is outlined. Applications of the strategy are reported that deal with two series of compounds. Examples of toxicity and persistency models are illustrated.
Quantitative Structure-Activity Relationships (QSARs): An integrated multivariate approach for risk assessment studies
1990
Abstract
The conditions and methods for constructing reliable QSARs are revised in relation to each component of a QSAR study: the selection of a training set out of a QSAR compatible series, the collection of data pertinent to the descriptors matrix (X) and to the effects matrix (Y), the analysis of data to connect X to Y by a regression model, and the validation of the model. In discussing these conditions, attention is given to the constraints that arise from the theoretical foundation of QSARs as analogy models of local validity and to the complexity and limited knowledge about the mechanisms of action. Hence, emphasis is placed on the need and importance to adopt multivariate methods for dealing with (1) the characterization of the structures, (2) the selection of a representative set of training compounds, and (3) analysis of the data. It is finally shown that the same integrated multivariate approach applies to the modeling of biological activities and other properties - chemical and biological - as well as to the modeling of correlations between batteries of data. The role of QSAR in risk assessment is addressed in the second part of the article. The framework of a strategy for an efficient screening assessment of toxic substances through the modeling of their exposure and toxicity-related properties is outlined. Applications of the strategy are reported that deal with two series of compounds. Examples of toxicity and persistency models are illustrated.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.