Abstract

ABCB1 belongs to a superfamily of ATP-dependent, membrane embedded, multispecific efflux pumps. It is noted for its abbility to confer resistence to therapeutical chemotherapy, and made responsible for poor permeation properties of drugs in the gastrointestine and in the blood-brain barrier. Resistance to a broad variety of chemotherapeutic agents (MDR, multi drug resistance) is mediated by ABCB1 by the mechanism of overexpression of the transporter in tumour cells. Therefore, inhibitors of ABCB1 are proposed to resensitise multidrug resistant cancer cells. Unfortunately, due to a sort of promiscuity in the binding interaction of ABCB1 with ligands, the use of rational drug design approaches results rather difficult and general applicaple in silico models are still rare. We generated predictive QSAR models for ABCB1 inhibitors by applying novel similarity based methods on our in house database. The SIBAR approach uses Euclidean distances between dataset compounds and a set of reference compounds as input variables for PLS analysis. MIMIC and MiPhaK calculate 3D based similarity values by a molecular matching procedure (MIMIC) and by calculation of interaction energies between a target and a probe atom (MiPhaK). As demonstrated by our results, the choice of the reference set is of utmost importance for calculation of SIBAR-descriptors. Comparing the predictive capacities of the models, the different methods showed almost equal performance. SIBAR-descriptor calculation is computationally not as extensive as the other methods. Thus, for screens of huge databases SIBAR is probably the most adequate method. Additionally, a ABCB1 substrate model was established as a filtering tool in early phases of the drug development process. We present a model that combines data from literature with data based on a classification scheme by Szakács et al.