Structure-Based Virtual Screening Using Index Technology

The standard approach to structure-based high-throughput virtual screening is a sequential procedure: Each molecule of a given library is evaluated against the target protein, eventually generating a ranked list of molecules. A new approach, TrixX BMI, avoiding the sequential screening pipeline is presented in this thesis. It is based on a novel descriptor that encodes physicochemical properties of small molecules and proteins. The descriptor is calculated as part of a preprocessing step for all molecules in a compound library. Compound flexibility is accounted for using pre-enumerated conformational ensembles and the descriptors are stored in an indexed database. Complementary site descriptors of the protein are used to identify matching compounds and possible placements within the active site. Chemical and shape complementarity is evaluated solely on the descriptor level. During this process, molecules that are not part of any descriptor match are discarded. This non-sequential access to the compound library results in a speed-up of one order of magnitude compared to competing approaches while yielding results of comparable quality. Furthermore, TrixX BMI can incorporate requests for a certain pharmacophore interaction pattern to the protein. In this scenario, the speed-up increases to two orders of magnitude. TrixX BMI can be deployed in a parallel computing environment and scales with the number of available cores. Thus, it is suited for virtual high-throughput experiments.