The QSAR Discriminant‐Regression Model
Oleg A. Raevsky, A. M. Sapegin, Николай Алексеевич Зефиров · Quantitative Structure-Activity Relationships · 1994
Abstract The DIscriminant‐REgression Model (DIREM) belongs to the ordinary discrete‐continuous QSAR approach, and has been proposed to explore SAR's for compounds with the same type of activity but different features of biological action. As a function scaling method, DIREM reveals the correspondence between structure descriptions and the distribution of compounds on a scale of biological activity. Model building is an iterative process that includes cluster formation using a stepwise procedure of linear discriminant analysis (LDA), and the discovery of quantitative relationships within each cluster by means of stepwise multiple regression analysis (MRA). DIREM is available in the software packages CLARAS (CLAssification and Recognition of Active Structure) and DNESTR (Design of New Effective STRuctures). Examples of DIREM application (guppy toxicity of aliphatic compounds, and β‐adrenergic activities of phenethylamines) are discussed.