Protein-Ligand Inverse Screening and its Application in Biotechnology and Pharmacology

The thesis at hand presents the development of a new computational target prediction method. Small molecules are rarely only binding to a single protein, but can interact with numerous proteins, their targets. Ignorance of molecule-protein interactions can lead to various problems, wherein the most dangerous probably are side-effects evoked by drugs binding to so-far unknown off-targets. Resolving these problems is the aim of target prediction methods, which try to find all target proteins for small molecules. The iRAISE method developed in this thesis faces the special requirements of structure-based inverse screening which in comparison to normal screening (one protein, many ligands) predicts targets for one small molecule from large protein libraries. In order to account for the large amounts of protein structural data, iRAISE introduces a database representation for efficient and consistent handling and storing of protein data. Further, protein active sites and small molecules are in the first screening step abstracted by a descriptor representation. The chosen descriptor contains features encoding the interaction pattern and the shape of the active site/molecule. Thus, by matching complementary descriptors, the need for sequential protein-ligand matching on atomic level is avoided. Inter-target ranking has been improved compared to standard protein-ligand scoring functions, which mostly contain a bias towards certain protein structures. A multi-step Scoring Cascade considers the reference ligand as well as the coverage of the ligand and pocket, and thus allows the scoring of structurally diverse pockets. Moreover, a Gaussian-based scoring assesses the significance of a score. Along with the new target prediction method, an evaluation strategy with new data sets has been developed. The evaluation of binding mode prediction, target ranking and running time shows promising results for iRAISE. Further, structure-based computational methods were successfully applied in a biotechnological case study of the development of a synthetic multi-enzyme pathway, highlighting the application potentials of these methods in areas besides drug design where their use is already established. Die vorliegende Arbeit präsentiert die Entwicklung einer neuen, computergestützen Methode zur Zielproteinvorhersage. Kleine organische Moleküle binden nur selten an ein einziges Protein, sondern interagieren mit einer Reihe von verschiedenen Proteinen, sogenannten Zielproteinen oder