HyPred predicts the density of solvent in the hydration layer around proteins. Given the hydration shell density crystallographic water sites can be predicted. A web server for doing HyPred calculations can be found here. A detailed description of HyPred can be found in [1]. A summary of HyPred can be found below.

A set of proximal radial distribution functions (pRDFs) are defined and calculated for as series of different atom types in proteins using all-atom, explicit solvent molecular dynamic simulations for three globular proteins. We evaluate solvent densities in the first few hydration layers and use these densities to generate a model for hydrated proteins without the need for running additional computationally expensive MD simulations. The model is constructed from the average over the MD simulation of the solvent's proton density profile surrounding the protein. Pettitt has perfomed similar work [2].

Each of the simulation boxes is partioned into cubes 0.5 Angstroms on a side. The density is evaluated at 1 ps intervals for every cube situated outside the protein. Figure 1 a illustrates this. Each cube exterior to the protein is assigned to an atom on the protein surface whose scaled van der Waals surface is closest to the center of the cube. Figure 1 c illustrates this. An important difference between Pettitt's work and the work presented here is that here the distance to the scaled van der Waals surface is used instead of the distance to the nucleus of the atom.
A, B) The average solvent density around Ub is obtained from MD simulations with the protein atoms fixed. An 8 Å grid spacing is shown in A) and C), but 0.5 Å is used in the calculations. C,D) Using these data, the pRDFs are calculated by identifying the nearest solute atom and distance to each grid element; for example, the oxygen atom (red, upper left) is the closest solute atom to 4 grid elements (denoted with lines). E,F) Solvent density calculated by reversing the mapping protocol using the pRDF calculated for the fine and coarse atom type definitions. R-values are listed. The color scale is asymmetric, and hence, noise tends to make the bulk solution appear blue. G) Solvent density calculated using the averaged pRDFs obtained from Mb and HEWL using the fine atom-type definition. Only one layer of cubes is shown but protein atoms within a 3 Å of the layer are shown. Thus some proteins atoms can be seen above the slab.


Protein atoms are grouped into classes using two different categorizations of atom types. One specification collects heavy atoms into groups according to their element type (e.g., C, N, O, S...), while hydrogen atoms are grouped together depending on the atom to which they are bonded (e.g., CH, NH, OH...). The other more detailed specification defines the atom groups according to both their elemental character and amino acid type. The more detailed set assigns each unique atom in each amino acid type as an atom type. The detailed set provides one source of improvement over Pettitt who classifies atoms only according to the element. When cubes are equidistant from atoms of the same type, the densities of the cubes are averaged. This gives the pRDFs. Sample pRDFs are shown in figure 1 d.

The reconstruction of the hydration shell density without additional MD simulations begins with the protein in the absence of water. The protein and surroundings are partitioned into a grid of cubes as in the mean-field approach of the previous subsection. Let r designate the distance between the center of cube i and the closest scaled van der Waals surface, say of atom type A. Each cube i outside of the protein is assigned the density gA(r) from the pRDF for the given protein atom type A, the separation r, and the atom type closest to the cube's center. The scale factor (0.53) is optimized by minimizing the sum of the R factors of the three proteins. Densities in cavities are set to zero. A sample prediction for ubiquitin is illustrated in figure 1 g. Figure 1 b shows the average density from the MD simulation.

[1] Virtanen, Jouko, Makowski, Lee, Sosnick, Tobin R., & Freed, Karl F. (2010) Modeling the hydration layer around proteins: HyPred Biophysical Journal 99:1611-1619
[2] Makarov, VA Andrews, BK & Pettitt, BM (1998) Reconstructing the protein-water interface Biopolymers 45:469-78