Optimization methods for virtual screening on novel computational architectures

The numerous virtual screening (VS) methods that are used today in drug discovery processes differ mainly by the way they model the receptor and/or ligand and by the approach to perform screening. All these methods have in common that they screen databases of chemical compounds containing up to millions of ligands i.e. ZINC database. Larger databases increase the chances of generating hits or leads, but the computational time needed for the calculations increases not only with the size of the database but also with the accuracy of the VS method and the model. Fast docking methods with atomic resolution require a few minutes per ligand, while molecular dynamics-based approaches still require hundreds or thousands of hours per ligand. Therefore, the limitations of VS predictions are directly related to a lack of computational resources, a major bottleneck that prevents the application of detailed, high-accuracy models to VS. The current increase in available computer power at low cost due to novel computational architectures would enhance considerably the performance of the different VS methods and the quality and quantity of the conclusions we can get from screening. In this review, we will discuss recent trends in modeling techniques which, in combination with novel hardware platforms, yield order-of-magnitude improvements in the processing speeds of VS methods. We show the state of the art of VS methods as applied with novel computational architectures and the current trends of advanced computing.