Effects Of Atomic Charges On The Computations Of Intermolecular Interactions

The computation of protein-ligand interactions is the central step in virtual screening for drug design. The most important part is the electrostatic energy which depends on the atomic charge values. However, the conventional charge models usually give much different charge values and remain various drawbacks. This study will analyze the details about how the charge values affect the electrostatic energy computation, in order to get useful information for improving the computational accuracy. This article contains the following parts.1, The relationships between the atomic charges and the equilibrium distance or the energy were studied for atom pairs of H, C, N, O, P, and S. It is found that electrostatic repulsion becomes the crucial part when the charge product exceeds a small amount near 0.01. For atom pairs with opposite charges, the equilibrium distance will decrease rapidly when the charge product decreases from 0 to -0.4, while van der Waals repulsion will resist the decreasing of the interatomic distance when the charge product beyond this region. It is also found that the minimum energy linearly depends on the charge product except for the inflection point or a short transition region. The analysis of the influences of charges to energies also expand to multiple atom pairs.2, The 60 protein-ligand complexes were computed using DOCK6 to compare the nine charge models including Hirshfeld, NPA, Mulliken, MK, CHELP, CHELPG, AM1, AM1-BCC and Gasteiger. Considering the best scored conformations, three models Hirshfeld, CHELPG and Gasteiger can get better success rate. Checking the docked conformations which are most close to the experimental structures, two models Hirshfeld and CHELPG are better. When dividing the 60 complexes into two sets of charged ligands and neutral ligands, the results from the two sets are similar to those from the whole 60 complexes. Then analyzing the correlation between the computed interaction energy and the experimental binding affinity, three models AMl-BCC, Hirshfeld and Gasteiger are better if comparing the docked conformations. It also shows that all charge models can get better results for the charged ligand complexes than the neutral ligand complexes. Among them, AMl-BCC can get excellent energy prediction for both sets. Further analysis reveals that the electrostatic energy values roughly present a linear correlation with the charge values. Two models Hirshfeld and Gasteiger get better docking and scoring results for their smaller atomic charge values. AMl-BCC provides mild charge values and also gets better scoring results.3, Three complexes metallo-enzyme(comt), nuclear hormone receptor(mr) and kinase(cdk2), and 3276 small molecules from DUD database were used to test the enrichment effect of active molecules for AM1, AMSOL, Gasteiger and MMFF94. For comt complex, four another charge models were also computed, including Hirshfeld, MK, NPA and Mulliken. The results show that AM1 and AMSOL can get higher enrichment factor at the top 1% screening range, while Gasteiger and MMFF94 is better at the top 5% range. The QM charges also show the superiority at a wider range from the top 1% to the top 20%. Analying the enrichment effect of screening based on the ROC curve, it is found that MMFF94 is the best one for comt and mr, while AM1 and Gasteiger is better for cdk2. Besides, Mulliken and NPA also get good results in the overall region for comt.This article provides detailed analyses for the effects of the charge values and the conventional charge models on the computations of intermolecular interactions. Useful data have been obtained for improving the computational accuracy. The assessment of these charge models also suggests proper choices for future application.