Docking experiments were performed with AutoDock4 and compared with docking experiments with AutoDock3. For each complex, 50 docking experiments were performed using the Lamarckian genetic algorithm with the default parameters from AutoDock3. A maximum of 25 million energy evaluations was applied for each experiment. The results were clustered using a tolerance of 2.0 Å.
In the HIV cross dockings, ligand flexibility was limited to 10 torsional degrees of freedom, picking torsions that allowed the fewest number of atoms to move (freezing the core of the molecule). Flexible docking was performed allowing three torsions to rotate in residue ARG8, in both the A and B chains. The structural water (water 301) was included in complexes that included this water in the crystallographic structure, and hydrogen atoms were added in geometry that allowed hydrogen bonding to the flaps.
We have also changed the default model for the unbound system in the current version of AutoDock. Our previous method calculated internal energies for an extended form of the molecule, mimicking a conformation that might be expected when fully solvated11 (link). Results from beta testers, however, showed that this protocol has severe limitations when used for virtual screening. In cases where the ligand is sterically crowded, the artificial force field used to drive the ligand into an extended conformation tends to lead to conformations with sub-optimal energy. When the difference is calculated between this unbound conformation and the bound conformation, it leads to artificially favorable predictions of the free energy of binding. In response to this problem, we have returned to the default model of assuming that the unbound conformation of the ligand is the same as the bound conformation. Other options in AutoDock allow the user to use an energy-minimized conformation of the ligand as the unbound model.
In the HIV cross dockings, ligand flexibility was limited to 10 torsional degrees of freedom, picking torsions that allowed the fewest number of atoms to move (freezing the core of the molecule). Flexible docking was performed allowing three torsions to rotate in residue ARG8, in both the A and B chains. The structural water (water 301) was included in complexes that included this water in the crystallographic structure, and hydrogen atoms were added in geometry that allowed hydrogen bonding to the flaps.
We have also changed the default model for the unbound system in the current version of AutoDock. Our previous method calculated internal energies for an extended form of the molecule, mimicking a conformation that might be expected when fully solvated11 (link). Results from beta testers, however, showed that this protocol has severe limitations when used for virtual screening. In cases where the ligand is sterically crowded, the artificial force field used to drive the ligand into an extended conformation tends to lead to conformations with sub-optimal energy. When the difference is calculated between this unbound conformation and the bound conformation, it leads to artificially favorable predictions of the free energy of binding. In response to this problem, we have returned to the default model of assuming that the unbound conformation of the ligand is the same as the bound conformation. Other options in AutoDock allow the user to use an energy-minimized conformation of the ligand as the unbound model.
