Abstract

Human ABC-transporters, which act as drug carriers, are notorious for their pivotal role in influencing the pharmacokinetic fate of a plethora of marketed drugs and also for their contribution to MDR, a leading cause of failure of anti-cancer pharmacotherapy in clinical practice. In-silico methods have gained a lot of acceptance in the last years with respect to understand the molecular triggers that drive biological activity of small molecules on the one hand but also with respect to support rational decision making in early phases of drug development on the other hand. In this thesis different machine learning algorithms (Rule-based Modelling, SVM, RandomForests) are employed to characterize proprietary and public data sets of ABC-Transporter substrates and non-substrates. From a pharmacological viewpoint the thesis will concentrate on ABCB1. From a methodological viewpoint the thesis concentrates on the assessment of different feature selection methods, descriptor development (extension of the SIBAR approach), and evaluation of distance-to-model (applicability domain) measurements. An additional focus is also the in-silico characterization of MDR-selective ("collateral sensitive") molecules by means of Network-like Similarity Graphs (NSGs).