Deep learning augmented protein mapping software to screen large compound libraries

Fragment based drug discovery starts with screening libraries of fragment-sized organic molecules for binding to the target protein. The fragments cluster at binding hot spots, the most important regions for drug discovery, and can be extended into larger and higher affinity ligands. The protein-mapping program FTMap is a computational analogue of fragment screening experiments. Acpharis has licensed the docking engine of FTMap and developed the ATLAS software as an updated version of the FTMap program. While ATLAS is a useful tool for identifying binding sites and predicting druggability, with proper development it can provide much more valuable characterization of both the binding site and the preferred fragments. The major goal of this proposal is to develop a software package based on ATLAS that, starting from the structure of a target protein, will be able to reliably screen very large virtual compound libraries for potential hits. To achieve this major goal we propose the following developments. Our first goal is to identify regions on the target protein that have preferences for binding specific functional groups and to identify a set of bound fragments that can be used as seeds for 2D and 3D screening. This will involve four steps. (1) Developing a higher accuracy scoring function to enable discrimination among different functional groups. (2) Obtaining generalized pharmacophore information by iterative mapping, where the initial mapping, indicating preferences for certain functional groups, will be followed by more focused mapping using probes containing similar functional groups; (3) designing basic and extended fragment libraries for the two steps of mapping; and (4) improving the functional characterization of the site by adding binding information from the PDB using a novel pocket similarity algorithm. Once extended pharmacophores are established, we plan to use ensembles of binding fragments as pseudo-compounds to seed a ligand-based shape-matching search method to screen large libraries of compounds based on molecular similarity. The traditional 2D similarity search will be modified to account for the additional 3D information provided by the mapping. This will enable screening larger libraries and will yield more specific results than the existing 2D ligand based tools. Once we have a set of potential ligand hits, we will perform template based ligand placement to produce a variety of possible poses, and to score the refined poses.