Statistical Analysis, Investigation, and Prediction of the Water Positions in the Binding Sites of Proteins

Abstract

Water molecules play a crucial role in biomolecular associations by mediating a hydrogen bond network or filling spaces with van der Waals interactions. Although current drug design technologies have taken water molecule interactions into account, their applications are still limited to their reliance on either excessive computer resources or a particular potential energy model. Here, we introduce a statistical method that is based on experimentally determined water molecules in the binding sites of high-resolution X-ray crystal structures to predict the potential hydration sites in the binding sites of crystal structures of interest. By clustering and analyzing the various interaction patterns of water molecules in the training data set, we derived a tetrahedron-water-cluster model based on a series of residue group triplets that form feature triangles of different shapes. In the tetrahedral-water-cluster model, a triplet of three polar atoms in the residue group triplet acts as the vertices of the bottom triangle of the tetrahedron, and the water molecule that interacts with these three polar atoms is set as the top vertex of the tetrahedron. By comparing the shapes of the bottom triangles in the training data set with the shape of the triangle in the residue group triplets in the crystal structure of interest, we can identify the bottom triangle that is most similar to the one in the residue group triplet of the crystal structure of interest. According to the tetrahedron-water-cluster model, the hydration site for the residue group triplet in the crystal structure of interest can be predicted based on the height of the tetrahedron that has the most similar bottom triangle in the training data set. A test set containing 193 crystal structures was used to evaluate model performance, and extensive comparison with the recently published program Dowser++ revealed that our model is at least as good at providing an accurate set of the potential hydration sites in crystal structures of interest.

Multi-Body Interactions in Molecular Docking Program Devised with Key Water Molecules in Protein Binding Sites

“…Generally, water molecules found in crystal structures contribute to the shape and the flexibility of the binding sites, mainly by mediating the formation of the hydrogen bonds between the proteins and their ligands [ 3 ]. However, the impact of the water molecules is often ignored directly in traditional molecular docking simulations.…”

Prediction of water positions in the binding sites of proteins based on collections of multi‐source heterogeneous atoms

“…The predicted results of Dowser++ using default experimental parameters are shown in Table S5. From the comparison results, the composite tetrahedron model had advantages over other programs in predicting potential hydration sites in crystal structures: (a) the composite tetrahedron model yielded a 98.25% success rate, which was higher than that of the tetrahedron‐water‐cluster model (96.89%) and Dowser++ (96.37%); (b) rates with RMSD values ≤1.50 and 2.00 Å for all the predicted sites were, respectively, 64.60% and 69.77% for Dowser++, while the corresponding rates were 71.04% and 83.66% for the composite tetrahedron model (Figure b); (c) especially for the 193 experiment‐determined water molecules (Xiao et al, ), the composite tetrahedron model achieved rates of 75.63% and 86.88% (Figure a) within the precision ranges of 1.50 and 2.00 Å, respectively, which were approximately equal to those of the tetrahedron‐water‐cluster model (about 73.26% and 87.70% within the precision ranges of 1.50 and 2.00 Å). However, the composite tetrahedron model was able to predict more potential sites of water molecules in the binding sites of proteins than the tetrahedron‐water‐cluster model.…”

“…The predictive performance of the composite tetrahedron model was further evaluated against the same test set (Xiao, He, Sun, Li, & Li, ) from two aspects: the success rate (the ratio of key water molecules that can be predicted from the dataset) and the root‐mean‐square deviations (RMSDs) between predicted sites of water molecules and experimentally determined water molecules in the crystal structures.…”

“…Only nine collections of multi‐source heterogeneous atoms of triplets were filtered out because the side lengths of the triangles exceeded the maximum thresholds of the corresponding side lengths. Especially for the 193 experimentally determined water molecules (Xiao et al, ), 33 triplets were filtered because their PF values were lower than the threshold, and the remaining potential hydration sites were all predicted by the composite tetrahedron model. With respect to RMSD values, the composite tetrahedron model achieved the success rates of 71.04 and 83.66% within the ranges of 1.50 and 2.00 Å, respectively (Figure ).…”

“…The detailed prediction process is as follows: To further develop and improve the prediction of water molecules, we used a training set of 2003 protein–ligand complexes (Xiao et al, ) collected from the Protein Database Bank (PDB). For each training complex, the water molecules within 7.0 Å of any ligand atoms (Garcia‐Sosa et al, ; Verdonk et al, ) were extracted as the research objects. According to the theory of atomic hybridization orbital, a water molecule had hydrogen bond interactions with four nearby polar atoms at most.…”