Conformation-dependent binding prediction

Is it possible to use descriptors of protein conformations to identify active ligands for that protein?

Jana is working in collaboration with Dr. Vineetha Menon's lab at UAH on a research project exploring protein conformations and their binding affinities. She is working with G-Protein Coupled Receptors (GPCR), due to their frequency as a pharmacological target. Jana is currently computing descriptors of binding site structure from a series of simulated protein conformations. These descriptive binding site properties will be analyzed using big data techniques. This methodology will be used to predict conformations that are most likely to show the best performance in docking simulations. Jana hopes to find a core set of protein and binding site descriptors that can be used to predict binding site affinities. This would allow a smaller subset of protein conformations to be tested in both docking simulations and laboratory settings. Jana’s research could lead to better predictions for drug-targeting protein conformations and provide exclusionary criteria to identify protein conformations with a high likelihood of off-target binding.

The image above is a close-up view of the binding pocket of ADORA2A human adenosine receptor bound to agonist UK-432097. This conformation of ADORA2A was determined to have selectively bound more known ligands through analysis of protein-ligand docking simulations.

The image to the right is one of the ADORA2a conformations, obtained through molecular dynamic simulations, superposed with the ADORA2A entry from the the Protein Data Bank (3QAK). The protein was modeled for 1 μs with snapshots saved every 200 ps. This is the conformation at 1400 ps, which was determined to be a favorable binding conformation.

The image above shows three consecutive snapshots of the ADORA2A protein taken at 1400, 1600 and 1800 ps, from left to right. Each is shown with the agonist UK-432097 in the binding pocket. The conformations at 1400 and 1800 (left and right) were determined to be favorable binding conformations during the analysis of the docking simulations. The 1600 conformation was not found to be favorable. Currently descriptors of the properties of both the entire protein conformation and the binding pocket are being developed to enable a analysis of the descriptors through machine learning methods to predict favorable binding conformations.