Abstract: Quantitative structure-activity relationships (QSARs) for 43 benzofuran and benzothiophene biphenyls were studied. By using a genetic algorithm (GA), a group of multiple regression models with high fitness scores (r2 was up to 0.70) were generated. From the statistical analyses of the descriptors used in the evolution procedure, four of them, including the partition coefficient (lgP), the molecular surface area (Area), the molecular weight (MW), and the dipole vector (Dip) were found to be the principal features affecting the biological activity. For example, the molecular surface area appeared in 94％ of the models in the elite populations. That is to say, the hydrophobic interactions between the inhibitors and the receptors are very important to the biological activity, which supplies a guide for the design and reconstruction of new PTP1B inhibitors.

Key words: 2D-QSAR, Genetic algorithm, PTP1B inhibitors