APPLICATION OF GENETIC FUNCTION APPROXIMATION TO QUANTITATIVE STRUCTURE-ACTIVITY-RELATIONSHIPS AND QUANTITATIVE STRUCTURE-PROPERTY RELATIONSHIPS

The genetic function approximation (GFA) algorithm offers a new approa ch to the problem of building quantitative structure-activity relation ship (QSAR) and quantitative structure-property relationship (QSPR) mo dels. Replacing regression analysis with the GFA algorithm allows the construction of models competitive with, or superior to, standard tech niques and makes available additional information not provided by othe r techniques. Unlike most other analysis algorithms, GFA provides the user with multiple models; the populations of models are created by ev olving random initial models using a genetic algorithm. GFA can build models using not only linear polynomials but also higher-order polynom ials, splines, and Gaussians. By using spline-based terms, GFA can per form a form of automatic outlier removal and classification. The GFA a lgorithm has been applied to three published data sets to demonstrate it is an effective tool for doing both QSAR and QSPR.