RESUMO
We propose an approach to rational design of new polyketides with the required spectrum of biological activity. We developed BioGenPharm software for generation of polyketide combinatorial libraries, prediction of activity spectra for the generated structures and selection of molecules with the required properties on the basis of user defined input parameters and selection criteria. For prediction of polyketide activity spectra we used PASS algorithm (http://www.ibmc.msk.ru/PASS). Validation of PASS prediction ability for polyketides was performed vs. the evaluation set containing 242 natural macrolides from the Dictionary of Natural Products. The mean prediction accuracy was 75,5%. The problem of choice of cutting points for probability of the presence of activity (Pa), which provide optimal combination of such parameters as sensitivity, specificity, concordance was considered. Applicability of the described method has been illustrated by generation of a virtual library of the erythromycin analogues and selection substances for which the probability of hepatotoxic action is low.
