Neighbor-Group Method And Descriptor Selection For QSAR Based On SVR

The quantitative structure-activity relationship(QSAR) is used in the field ofchemistry, and relates physico-chemical properties of chemical compounds to theirstructures It is well known for instance that within a particular family of chemicalcompounds, especially of organic chemistry, that there are strong correlations betweenstructure and observed properties.In many cases the approaches use curve fitting, interpolation and extrapolationtechniques. The predictor variables can be a variety of chemical descriptors. In thispaper, support vector regression(SVR) was applied to represent the QSARrelationships for substituted anilines and phenols to Daphnia magna Straus,sulfonanilide herbicides and sulfonylurea herbicides, by choosing optimal kernelfunction, screening descriptors, selecting optimal text. The results show thatneighbor-group method and descriptors selection based on SVM not only hold on finelearning ability, give better prediction performance and steady capability, but also hasthe low computation complexity. So these methods are expected to have broadprospect in evaluating and controlling environmentally toxic compounds.