A method of finding an informative set of molecular graph indices for the prediction of chemical compound properties

The method and algorithms for predicting properties of chemical compounds by their quantitative structural characteristics, in particular, molecular graph indices, are presented. The prediction procedure consists of establishing the priority of indices for training sample compounds, classifying control sample compounds in the Euclidean space of indices, and finding a locally optimum (informative) index set giving a least prediction error. The algorithms have been successfully tested using the BACC system (analysis and classification of biologically active compounds), created at the S. V. Sobolev Institute of Mathematics of the Siberian Branch of the Russian Academy of Sciences. Translated fromZhumal Strukturnoi Khimii, Vol. 38, No. 4, pp. 795–802, July–August, 1997.