Simulation of Reversible Protein–Protein Binding and Calculation of Binding Free Energies Using Perturbed Distance Restraints

Abstract

Virtually all biological processes depend on the interaction between proteins at some point. The correct prediction of biomolecular binding free-energies has many interesting applications in both basic and applied pharmaceutical research. While recent advances in the field of molecular dynamics (MD) simulations have proven the feasibility of the calculation of protein–protein binding free energies, the large conformational freedom of proteins and complex free energy landscapes of binding processes make such calculations a difficult task. Moreover, convergence and reversibility of resulting free-energy values remain poorly described. In this work, an easy-to-use, yet robust approach for the calculation of standard-state protein–protein binding free energies using perturbed distance restraints is described. In the binding process the conformations of the proteins were restrained, as suggested earlier. Two approaches to avoid end-state problems upon release of the conformational restraints were compared. The method was evaluated by practical application to a small model complex of ubiquitin and the very flexible ubiquitin-binding domain of human DNA polymerase ι (UBM2). All computed free energy differences were closely monitored for convergence, and the calculated binding free energies had a mean unsigned deviation of only 1.4 or 2.5 kJ·mol–1 from experimental values. Statistical error estimates were in the order of thermal noise. We conclude that the presented method has promising potential for broad applicability to quantitatively describe protein–protein and various other kinds of complex formation.

Correcting electrostatic artifacts due to net‐charge changes in the calculation of ligand binding free energies

“…One of the major challenges in modern computational biochemistry is the accurate and reliable calculation of affinities between proteins and associated binding partners . A very prominent method to calculate free‐energy changes upon a ligand binding to a host protein or DNA is to simulate a nonphysical path where the ligand is alchemically perturbed in the protein and free in solution .…”

“…One of the major challenges in modern computational biochemistry is the accurate and reliable calculation of affinities between proteins and associated binding partners. [1][2][3][4][5][6][7][8] A very prominent method to calculate free-energy changes upon a ligand binding to a host protein or DNA is to simulate a nonphysical path where the ligand is alchemically perturbed in the protein and free in solution. [9] In combination with the employment of a thermodynamic cycle, the free-energy change along the nonphysical path is the same as the free-energy change along the physical path.…”

Specific Residue Interactions Regulate the Binding of Dengue Antigens to Broadly Neutralizing EDE Antibodies

“…The method produces comparable results to alternative methods that pursue free energy calculations based on a geometrical route . Other methods were able to achieve more full transitions from the unbound state to the bound state in previous studies . These methods efficiently performed the enhanced sampling, but they were applied to other protein systems different in molecular size from the proteins considered in this study.…”

“…[16,17,35,38,39] Other methods were able to achieve more full transitions from the unbound state to the bound state in previous studies. [40,41] These methods efficiently performed the enhanced sampling, but they were applied to other protein systems different in molecular size from the proteins considered in this study.…”

iSEE: Interface structure, evolution, and energy‐based machine learning predictor of binding affinity changes upon mutations

“…Computational methods for ΔΔG prediction can be largely grouped into three main strategies: (1) Rigorous methods, such as thermodynamic integration and free energy perturbation, 4,5 (2) empirical energy-based methods, based for example on classical mechanics or statistical potentials [6][7][8][9][10] (typically in linear forms), and (3) machine learning-based methods which can exploit a large variety of energetics and non-energetics (eg, geometric, evolutionary) features. [11][12][13] The rigorous methods can be accurate but they are computationally highly demanding.…”