We compared two docking methods for their ability to estimate Ca2+
-protein affinities, Fold-X9
and AutoDock vina.10
Fold-X has been published as a method for the prediction of positions of metal-ions on proteins, and for the prediction of affinities between proteins and metal-ions. The energetic model underlying Fold-X has an ad hoc
form with a number of parameters that have been fitted to experimental data.9
We obtained Fold-X versions 2.5.2 and 3.0b3 as executables for Linux from the respective server, including documentation. Despite considerable efforts we were not able to generate estimates of affinities of metal-ions to proteins using the commands described in the documentation of the software; we suspect that the option in those versions of the software is dysfunctional. Fortunately, the authors of Fold-X have offered as Table 4 in their supplementary material to Ref9
a list of 48 Ca2+
-binding pockets in 19 X-ray structures, mostly with experimentally determined affinities, and affinities predicted with an earlier version of Fold-X. Hence, we took these data (“Fold-X dataset”) as basis for the comparison, specifically the columns “experimental energy” and “predicted energy” of Table 4 in Ref.9
For five of the binding pockets, two experimental energies were given; since the differences between the first and second energies were relatively small, only the first value was considered in each case. For ease of comparison with Ref 9
we give all affinities in units of kcal/mol (1 kcal/mol = 4.1868 kJ/mol).
The 19 X-ray structures of the Fold-X dataset were retrieved from the Protein Data-bank (PDB4
) for comparative analysis with AutoDock vina.10
AutoDock vina version 1.0.3 for Linux was downloaded as executable from the website of its authors.
Usually, some information is missing from X-ray structures that is needed for energy calculations. Most importantly, this is the case for hydrogen positions, including also hydrogen bond networks. Related to this is the possibility to optimize X-ray structures by flipping carbonyl-oxygens and –NH groups (both groups have similar electron densities). Finally, sometimes X-ray structures contain atomic overlaps that can be removed relatively easily. To see whether by considering these effects, affinity predictions can be improved, we used three different protocols with AutoDock vina.
In the first protocol, PDB files were prepared with the AutoDockTool suite of AutoDock 411
by removing water molecules, non-standard residues, and alternate positions of residues. Then polar hydrogens were added to the protein in standard orientation without rotational optimization, and Gasteiger charges12
were computed for protein atoms by AutoDock. Finally, all Ca2+
-ions in the protein were redocked with AutoDock vina using default parameters, except for the “search space”, ie, the volume in which the optimal docking position is searched for. Since we were interested in affinities at the crystallographically determined positions of the Ca2+
-ions, and not in finding optimal positions, we restricted the search space to the minimum allowed by AutoDock vina around the crystallographically determined positions, namely a cube of 1 Å (0.1 nm) length in x-, y-, and z-directions around the crystallographic positions of the Ca2+
-ions. This protocol was applied to all Ca2+
-ions in the Fold-X dataset.
In the second protocol we introduced in the preparation of the protein structures a further step in which the “reduce” method13
was used to optimize positions of hydrogen atoms around crystallographically determined heavy-atom positions, including also potential flips of amide-groups in side chains of asparagine and glutamine. Otherwise this second protocol had the same elements as the first, including also the restrained docking of Ca2+
-ions with AutoDock vina.
The third protocol was similar to the first one, but additionally optimized hydrogen positions with pdb2pqr,14
including debumping to avoid steric clashes and optimization of the hydrogen bonding network.
For computations of solvent accessible surfaces (SAS) we used MSMS version 2.5.715
with standard atomic radii for protein atoms and 1 Å for Ca2+
. All statistical analyses were carried out with R.16