Advanced Computational Methodologies to Study Binding Free Energies of Biomolecular Complexes

Molecular recognition is the basis of biological mechanisms and is a key element to consider while formulating effective and safe drugs. Pharmaceutical drugs are designed so as to bind a target protein even at very low concentrations to alter the diseased conditions without interfering with normal biological processes. In a rational drug design process, this is achieved by acquiring information about the chemical structure and the physical and chemical properties of the target protein receptor to gain insights on how changing the chemical composition of the substrate drug could affect the protein-drug interaction and binding affinities. Computational models are used in conjunction with traditional experimental techniques to pre-screen lead drug-like molecules before they are synthesized and assayed. Often, due to their computational speed, empirical molecular docking algorithms are used for these purposes as opposed to more accurate physics-based models. Computational physics based models are also more difficult to properly set up and require long simulation times to give reliable estimates of binding affinities to be effectively used for virtual screening purposes. In this work, new methodologies have been developed to improve on critical factors that affect the accuracy and usability of physics-based alchemical free energy models with the overarching goal to eventually make these models applicable to free energy based virtual drug screening. The critical factors that are addressed in this work are as follows: (i) the limitation of implicit solvent model in proper treatment of hydration is addressed by developing a hybrid solvent model, combining the thermodynamic properties of water molecules obtained from explicit solvent simulations with the implicit solvent energy function. The hybrid approach significantly improved the binding affinity predictions in different systems tested, including host-guest and Dopamine D3 complexes, (ii) to properly account for conformational reorganization effects in simulations, improved force-field parameters are implemented for reliable prediction of Farnesoid X receptor complexes, (iii) insufficient conformational sampling resulting from entropic bottlenecks in the alchemical transformations are avoided by implementing better soft-core potentials and introducing a customized alchemical perturbation schedule. These together have been shown to be effective in reducing the strength of a rare order/disorder phase transitions in alchemical simulations of binding, allowed interconversion between different binding modes of complexes, and achieve better convergence of binding free energy estimates of protein-ligand complexes studied in this work