Abstract:

Computational science has the potential to solve most of the problems which pharmaceutical research is facing these days. In this field the most pivotal property is arguably the free energy of binding. Yet methods to predict this quantity with sufficient accuracy, reliability and efficiency remain elusive, and are thus not yet able to replace experimental determinations, which remains one of the unattained holy grails of computer-aided drug design (CADD). The situation is similar for methods which are used to identify promising new drug candidates with high binding affinity, which resembles a closely related endeavor in this field. In this thesis the development of a new free energy method (QSTAR) was in the focus. It is able to explicitly take into account the quantum nature of atomic nuclei which so far was not done in binding free energy simulations of biomolecular systems. However, it can be expected to play a substantial role in such systems in particular due to the abundance of hydrogen atoms which posses one of the strongest nuclear delocalizations of all atoms. To take these nuclear quantum effects into account Feynman’s path integral formulation is used and combined in a synergistic way with a novel alchemical transformation scheme. QSTAR makes also available the first readily available single topology approach for electronic structure methods (ESMs). Moreover, an extended alchemical scheme for relative binding free energies was developed to address van der Waals endpoint problems. QSTAR and the alchemical schemes were implemented in HyperQ, a new free energy simulation suite which is highly automated and scalable. Most ESMs methods become soon prohibitively expensive with the size of the system, a restriction which can be circumvented by quantum mechanics/molecular mechanics (QM/MM) methods. In order to be able to apply QSTAR together with ESMs on biomolecular systems an enhanced QM/MM scheme was developed. It is a method for diffusive systems based on restraining potentials, and allows to define QM regions of customizable shape while being computationally fast. It was implemented in a novel client for i-PI, and together with HyperQ allows to carry out free energy simulations of biomolecular systems with potentials of very high accuracy. One of the most promising ways to identify new hit compounds in CADD is provided by structure- based virtual screenings (SBVSs) which make use of free energy methods. In this thesis it is argued that the larger the scale of virtual screenings the higher their success. And a novel workflow system was developed called Virtual Flow, allowing to carry out SBVS-related tasks on computer clusters with virtually perfect scaling behavior and no practically relevant bounds regarding the number of nodes/CPUs. Two versions were implemented, VFLP and VFVS, dedicated to the preparation of large ligand databases and for carrying out the SBVS procedure itself. As a primary application of the new methods and software a dedicated drug design project was started involving three regions on the novel target EBP1, expected to be located on protein-protein interfaces which are extremely challenging to inhibit. Three multistage SBVSs were carried out each involving more than 100 million compounds. Subsequent experimental binding assays indicated a remarkably high true hit rate of above 30 %. Subsequent fluorescence microscopy of one selected compound exhibited favorable biological activities in cancer cells. Other applied projects included computational hit and lead discovery for several other types of anti- cancer drugs, anti-Herpes medications, as well as antibacterials.