Article ID: 1034 - Last Modified: May 4, 2012
Do Glide docked poses depend on the initial ligand conformations?
Complex numerical algorithms, such as those in Glide, can exhibit sensitivity to initial conditions, so some variation in the poses and scores is expected when starting from different input conformations of a particular ligand. This is particularly true when the bond lengths and angles changes, but is also true for input conformations with only torsional differences, or even to input structures of the same ligand conformation that differ in absolute coordinates.
Sampling, docking, and scoring in a timely manner is a challenging problem. Dependence on the input conformation arises because of the complex grid-based potentials and the practical limitations on thoroughly sampling the docked poses. In addition, Glide generates new ligand conformations through torsional variations only, so any differences in bond lengths and angles in the input ligand structures will persist through the docked poses, resulting in scoring and pose differences.
In general, ligands that are good binders will fluctuate less than poor binders. A poor binder or inactive is less likely to have a low energy solution that can be found independent of small variations. Also, poor binders are more likely to encounter the penalty terms in Glide XP, and these penalties can vary significantly for small changes in conformation. However, even a good binder can exhibit fluctuations in poses and scores if it is a difficult sampling situation, such as for a tight binding pocket or a large ligand.
One option for eliminating these input dependencies is to use the "Regularize input geometries" feature of the Virtual Screening Workflow (on the Preparation tab). This converts the input ligands to unique SMILES and back to 3D, so that a given ligand will always produce the same result, regardless of input geometry or coordinates.
A way to reduce the input dependence that arises from the torsion-only conformation search is to dock several conformations of each ligand with variations in bond lengths and bond angles. The XP docking stage of the Virtual Screening Workflow has an option to 'Generate multiple input conformations', which produces extra input conformations for each ligand by running MacroModel calculations with two different force fields (MMFFs and OPLS_2001).
Type the words or phrases on which you would like to search, or click here to view a list of all
Knowledge Base articles