Assessing the performance of MM/PBSA and MM/GBSA methods. 5. Improved docking performance using high solute dielectric constant MM/GBSA and MM/PBSA rescoring

With the rapid development of computational techniques and hardware, more rigorous and precise theoretical models have been used to predict the binding affinities of a large number of small molecules to biomolecules. By employing continuum solvation models, the MM/GBSA and MM/PBSA methodologies achieve a good balance between low computational cost and reasonable prediction accuracy. In this study, we have thoroughly investigated the effects of interior dielectric constant, molecular dynamics (MD) simulations, and the number of top-scored docking poses on the performance of the MM/GBSA and MM/PBSA rescoring of docking poses for three tyrosine kinases, including ABL, ALK, and BRAF. Overall, the MM/PBSA and MM/GBSA rescoring achieved comparative accuracies based on a relatively higher solute (or interior) dielectric constant (i.e. ε = 2, or 4), and could markedly improve the 'screening power' and 'ranking power' given by Autodock. Moreover, with a relatively higher solute dielectric constant, the MM/PBSA or MM/GBSA rescoring based on the best scored docking poses and the multiple top-scored docking poses gave similar predictions, implying that much computational cost can be saved by considering the best scored docking poses only. Besides, compared with the rescoring based on the minimized structures, the rescoring based on the MD simulations might not be completely necessary due to its negligible impact on the docking performance. Considering the much higher computational demand of MM/PBSA, MM/GBSA with a high solute dielectric constant (ε = 2 or 4) is recommended for the virtual screening of tyrosine kinases.