Skip to main content
Open Access Publications from the University of California

Predicting charged protein-ligand binding affinities using free energy calculations

  • Author(s): Rocklin, Gabriel Jacob
  • Advisor(s): Shoichet, Brian K
  • Dill, Ken A
  • et al.

Predicting protein-ligand binding free energy from physical principles is a grand challenge in biophysics, with particular importance for drug discovery. Free energy calculations compute binding affinities by using classical mechanics to model the protein and ligand at atomic resolution, and using statistical mechanics to analyze simulations of these models. The binding affinities computed from these simulations are fully rigorous and thermodynamically correct for the model (with adequate sampling), and will agree with experimentally measured binding affinities if the model is accurate. Because free energy calculations capture the full statistical complexity of binding for flexible molecules at ambient temperature, they offer the greatest potential for quantitative accuracy of any physical method for predicting binding.

Here, I (& coauthors) present several studies relating to using free energy calculations to predict protein-ligand binding affinities for charged compounds. First, we introduce the Separated Topologies method, an approach for using free energy calculations to predict relative binding affinities of unrelated ligands. This method is useful for studying charged compounds because charged compounds are very difficult to study using absolute binding calculations, increasing the importance of relative binding calculations. Second, we use free energy calculations to predict absolute binding affinities for charged molecules to a simplified protein binding site, which is specially designed for studying charged interactions. These predictions are compared to new experimental affinity measurements and new high-resolution structures of the protein-ligand complexes. We find that all affinities are predicted to be too strong, and that this error is directly correlated with the polarity of each ligand. By uniformly weakening the strength of electrostatic interactions, we are more successful at predicting binding affinity. Third, we design and validate an analytical correction scheme to correct binding free energy calculations of ions for artifacts caused by the periodic boundary conditions employed in simulations. Fourth, we examine the sensitivity of binding affinities from free energy calculations to the force field parameters used in the simulations. This provides insight into the strength of electrostatic interactions in protein simulations, complementing our previous work comparing simulation results to experiments. Finally, we discuss potential future directions of this work.

Main Content
Current View