ИСТИНА

The Molecular Field Topology Analysis (MFTA) technique [1] is intended to model the relationships between the biological activity of compounds and the parameters of their atoms and bonds. Meaningful comparison of these local molecular properties is enabled by a so-called molecular supergraph such that any training set structure can be superimposed onto it. To build a uniform descriptor set, each supergraph position is assigned a property value for a corresponding atom or bond in a structure (e.g., effective atomic charge, van der Waals radius, H-bond donor and H-bond acceptor ability, local lipophilicity and/or other parameters) while for unoccupied vertices the neutral descriptor values are used. The statistical analysis of this descriptor set (usually by means of the Partial Least Squares Regression) yields a predictive model that allows one not only to predict the bioactivity for new compounds but also to identify the structural features critical for activity (i.e., related to the local descriptors making the largest contribution to the activity). Thus, the MFTA modelling provides a useful tool for the targeted virtual screening [2] of novel promising structures. This approach has been successfully applied to many chemical classes and kinds of bioactivity, ranging from antiviral agents to neuroreceptor ligands to irreversible inhibitors of serine esterases. However, the experience indicates some areas where improvement is desirable. In its original form, MFTA is best suited for series of congeneric structures involving a single common core and various peripheral fragments. If such ‘substituent’ fragments are too complex and diverse, the intuitively desirable superposition of structures may be problematic to achieve and/or significant manual intervention may be required. In addition, the ability to handle structurally different but roughly bioisosteric core fragments would be useful both to expand the applicable training set and to support predictions for related compounds as well as scaffold hopping. In view of these goals, the MFTA technique was extended to detect reasonable core fragments (e.g., polycyclic) that may be separated by more flexible linkers. The automatic selection of the cores may be adjusted by a researcher, affecting further superimposition steps. Similarly, during prediction procedure new cores may be introduced and mapped to some of the already analyzed ones. The extended multi-core MFTA approach was implemented in the C++ software. Its application to a number of test cases shows promising results that are presented in the paper. [1] Palyulin V.A., Radchenko E.V., Zefirov N.S. J. Chem. Inf. Comput. Sci., 2000, 40, 659-667. [2] Radchenko E.V., Palyulin V.A., Zefirov N.S., in Chemoinformatics Approaches to Virtual Screening, ed. by A.Varnek, A.Tropsha, RSC, 2008, 150-181.