Undirected Exploration of Binding Pockets with Flexible Topology.

A common first step in drug design is virtual high-throughput screening (VHTS), where a large number of potential drug molecules are computationally modeled in a protein binding pocket and filtered down to a smaller set of hits that can be further tested computationally or experimentally. Traditional strategies for VHTS do not account for ligand-induced conformational changes in proteins as they typically rely on a single static structure to represent the protein. This neglects the role of binding entropy and the fact that different ligand molecules can induce slightly different conformations in the protein binding site that significantly affect the assessment of a given molecule's fit. To address this challenge, we have developed a method called "flexible topology", where a subset of atoms, typically representing a small molecule ligand, can continuously change their atomic identities, which are encoded by a set of attributes that parametrize the nonbonded interactions. These attributes are all implemented as dynamic variables that have masses and evolve over time using gradients of the energy function. In other words, the attributes feel forces from their surrounding environment and respond accordingly. In this way, by observing a set of flexible topology particles move and change in a ligand-binding site, we can learn the preferences of a binding pocket. Here, we demonstrate how undirected flexible topology simulations can be used to explore ligand-binding sites and reveal the desirable properties of potential ligands. We use the β-2-adrenergic receptor as an illustrative example and compare the properties of flexible topology particle groups with a set of 29 B2AR ligand-bound crystal structures, covering 13 distinct ligands. We also show how the shape- and electrostatics-based virtual screening software "eon" from OpenEye can be used to find hits that come as close as possible to mimicking the orientation of our flexible topology atoms.