The workflow will consist of a preliminary part and two production parts. On the preliminary stage MD simulation of the protein will be performed to study protein flexibility and choose representative conformations. On the first stage we will enumerate possible ligands for each of the representative protein conformations by the fragment-based de novo approach CReM guided by docking (Autodock Vina). To create the library of fragments used for compound enumeration we will use Enamine database of previously synthesized compounds that should increase synthetic accessibility of designed molecules. Hit lists obtained for every representative protein conformation will be combined in one and compounds will be filtered by physicochemical properties regarding the traffic-lights evaluation protocol, structural alerts, synthetic accessibility score and further by a retrosyntetic tool (AiZynthFinder). The remaining compounds will be redocked by other docking programs (smina, gnina, DeepDock, etc) and ranked independently by each scoring function. We will count the number of times each compound will get in top 1% in every ranking list and choose those compounds which occurred more frequently in top 1% of different lists (consensus docking). For selected compounds we will perform MM-PBSA calculation of binding free energy (rescoring) to remove less promising molecules. From the remaining compounds we will select 50 diverse molecules. On the second stage we will use these 50 de novo generated molecules as queries to perform similarity search among Enamine REAL space using FTrees or using similarity search among previously synthesized Enamine compounds. All selected compounds will be subjected to the same filtering, consensus docking and rescoring procedure as described above to select other 50 compounds, which should be easy to synthesize. Using such a two-stage pipeline we will be able to perform adaptive search in Enamine REAL space.
