Accurate and Efficient Calculation of Protein–Protein Binding Free Energy-Interaction Entropy with Residue Type-Specific Dielectric Constants

Abstract

Accurate and efficient computation of protein−protein binding free energy remains a grand challenge. In this study, we develop a new strategy to achieve efficient calculation for total protein−protein binding free energies with improved accuracy. The new method combines the recently developed interaction entropy method for efficient computation of entropic change together with the use of residue type-specific dielectric constants in the framework of MM/GBSA to achieve optimal result for protein−protein binding free energies. The new strategy is shown to be computationally efficient and accurate than that using standard MM/GBSA methods in which the entropic computation is performed by the normal model approach and the protein interior is represented by the standard dielectric constant (typically set to 1), both in terms of accuracy and computational efficiency. Our study using the new strategy on a set of randomly selected 20 protein−protein binding systems produced an optimal dielectric constant of 2.7 for charged residues and 1.1 for noncharged residues. Using this new strategy, the mean absolute error in computed binding free energies for these 20 selected protein−protein systems is significantly reduced by more than 3-fold while the computational cost is reduced by more than 2 orders of magnitude, compared to the result using standard MM/GBSA method with the normal mode approach. A similar improvement in accuracy is confirmed for a test set consisting of 10 protein−protein systems.

Computational prediction of protein–protein binding affinities

“…For example, Gohlke et al investigated the Ras–Raf and Ras–RalGDS complex and reported a binding free energy in good agreement with experiment, however, depending on how conformational entropy was estimated and with errors of about several kcal/mol. MM‐GBSA and MM‐PBSA were successfully used in several studies for reranking protein–protein docking solutions . A very systematic study to predict the free energy of binding and to score docked complexes was recently performed by Chen et al The authors compared various force fields, protocols for performing MD simulations and using the Poisson–Boltzmann or the GB solvation model (not including conformational entropy effects) for 46 protein–protein complexes.…”

“…Hence, further testing could be useful to check the robustness of the IE approach. In a study of 20 protein–protein systems using IE with MM/GBSA, the mean absolute error to experimental binding affinities could be substantially reduced by optimizing the residue type‐specific dielectric constants, the errors were especially lower than with NM analysis using a standard dielectric constant of 1 …”

The effect of multiple simulation parameters on MM/PBSA performance for binding affinity prediction of CB1 cannabinoid receptor agonists and antagonists

“…Studies attempting to isolate the effects of entropy calculations using normal mode analysis on MM/ PBSA performance have previously reported mixed results (Kongsted & Ryde, 2009;Su et al, 2015;Sun et al, 2018). The introduction of the interaction entropy method has however significantly increased the number of studies considering entropic contributions due to a combination of its theoretical rigour and the simplicity of its implementation, with many recent studies demonstrating significant improvements in MM/PBSA performance using this approach (Liu, Peng, & Zhang, 2019;Sun et al, 2018). While one study has demonstrated that utilizing this approach results in an overestimation of the entropic penalty for charged ligands and thus a reduction in overall performance (Suárez & Díaz, 2019), in our study, we observed no clear trends on potential factors leading to the observed reduction in performance.…”

Reliable in silico ranking of engineered therapeutic TCR binding affinities with MMPB/GBSA

“…"# values greater than 1 to greatly improve the quality of prediction for the exchange of charged residues. [16,[43][44][45][46] To assess the possibility that the outliers observed in the MMPBSA calculations with an ! "# of 1 were induced by changes in the charge of the TCR, we colored variants in Fig 2C+D according to the total number of charged mutations made from the WT.…”

“…refs. [6,[11][12][13][14][15][16]). With this in mind, we aimed to identify an MMPB/GBSA protocol that provides reliable and accurate relative binding free energies for a PPI of great interest in the field of immuno-oncology, [17] the T-cell receptor (TCR) peptide-human leukocyte antigen (pHLA) complex (TCR-pHLA, Fig 1).…”