It has been estimated there are around 10<sup>60</sup> potential organic molecules with MW &lt; 500, considering only basic structural rules, yet typical virtual screening decks include only a small fraction of this chemical space.  Beginning to bridge this gap are on-demand synthesizable screening libraries and synthetically tractable enumerated libraries on the order of hundreds of millions to low billions of compounds.  To virtually screen such massive libraries of compounds within project timelines of drug discovery projects, GPU-based Shape screening methods have demonstrated considerable utility. Hoping to take advantage of this, we have rewritten our shape screening algorithm as a CUDA application utilizing the Thrust library.  Screening timings, technical advantages, and utility of the method in drug discovery are demonstrated against the DUD-E dataset, investigating the impact of probe selection, the number of conformers employed in screening, and the impact of minimized or unminimized conformations.  Results from screens performed against hundreds of millions of compounds in the Enamine REAL dataset and a newly developed approach to synthetically-tractable targeted enumeration are examined.  Guidance is provided for applying the methdology alone in hit identification or as part of a multi-stage workflow where Shape screening is employed prior to structure-based virtual screening with Glide.

Challenges and opportunities for GPU-enabled shape screening of large scale, synthetically-tractable compound libraries

256th National Meeting (2018)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.